It is a commonplace that language as “an elaborated metarepresentational device” (de Saussure 2013: 68) is closely interrelated with cognitive capacities such as perspective-taking and Theory of Mind, and various studies have argued for a correlation between linguistic abilities of propositional attitude ascriptions and the performance on ToM tasks such as the false-belief test (cf. for overviews Astington & Baird 2005a; Jacques & Zelazo 2005: 144–146; Paynter & Peterson 2010). While it is controversial which capacities exactly can be paralleled with ToM capabilities, there is a close link between the structural components of ToM tasks and syntactic embedding under non-factive mental state verbs. This parallelism is already indicated in the definition by Dunbar (2006: 172):

The analogy seems quite obvious since complement clauses like the one in (1)

seem to reflect ToM as the ability to imagine another individual’s mental state insofar as the embedding structure of two mental state predicates puts two different subjects of consciousness on stage (i.e. I believe vs. you think). This suggests that there is a viewpoint switch from the speaker-I to another person’s stance, and, hence, perspective-taking. However, there are two aspects that go beyond a perspective switch. First, taking another person’s stance does not mean to take another person’s position and forget about one’s own. Rather, it is ‘taken’ while maintaining the relation to one’s own state of belief. The primary viewpoint and the second viewpoint are hence not equivalent options, but display a qualitative difference whereby one viewpoint is subordinated to the other. Second, both primary and secondary viewpoints are based on the complement structure of propositional attitude ascriptions: non-factive verbs of perception, cognition, or communication in the matrix clause indicate the perspective on the event, while the embedded clause contains the perspectivized entity (cf. Astington & Baird 2005b: 165; Verhagen 2005: 78). This is seen in the fact that utterances like (2) can be evaluated with respect to two different veridicality attributions but that only (2b) would be a natural reaction in discourse.

This complement structure is yet not a given. When children start producing sentences like I think p between the age of 2 and 3 years, mental state predicates are first used like epistemic stance markers (Diessel & Tomasello 2001). The subordination of the complement under the main clause is thus a later development. As Verhagen argues, it is only when this ability has developed that “it also becomes possible for a conceptualizer, in uttering I think, to construe his own thinking as an object of conceptualization for specific purposes, as in I think he will arrive on time, but I am not sure/but John is skeptical […]” (Verhagen 2007: 71). This is a crucial step since it allows for the detachment of one’s own viewpoint and to consider one’s own perspective (i.e. the relation between the origo and the perspectivized entity) as an object of evaluation. As a result, the own viewpoint becomes the origo and (part of) the object of evaluation at the same time. This recursive structure calls for a three-point concept of perspectivization, since the evaluation of the own viewpoint presupposes the awareness that there are other possible perspectives available, in other words: that p [the world is flat] could be false. According to de Villiers et al. (2014: 229), it is thus the recognition that the complement is false and hence the contrast of truth-values between clauses that triggers the recursion of complementation structures.

Note that the observations so far hold for first order sentences like in (2) which refer to one subject of consciousness only. In (1), the structure is even more complex since it involves the embedding of another person’s mental state and thus multiplies the relations between the different viewpoints. There is (i) a potential contrast between the two speaker beliefs (I believe vs. you think), and (ii) a potential contrast with respect to the embedded proposition that can be either true or false. While both contrasts are relevant with respect to the utterance, they do not display the same perspectival constellation. Contrast (i) between The world is flat vs. The world is not flat refers to two possible options that are structurally equal, but contradict each other in that only one of them can be true at the same time (see also Dancygier & Sweetser 2005; Dancygier 2012b for modelling the concept of ‘alternativity’ in mental space theory). The difference between I believe and you believe, on the other hand, relies on a contrast, but not a contradiction, as trivially, I can believe something other than you do – in the same way as it may be raining in Paris and not raining in New York. According to Perner et al. (2003), ‘real’ perspective problems are hence defined by the fact that the different viewpoints cannot be aligned by coordination.

The concept of perspective-taking as defined in Section 3 presupposes that there are (at least) two alternative equal viewpoints to choose from. That is to say, there is a choice between either option A or option B, as given in (3a). The structure in (3b) integrates both the contradiction on the propositional content level as well as the contrast between the two states of mind which shows that embedding structures can integrate more than one viewpoint at the same time. This becomes clear by the fact that the second viewpoint can be filled out by an external ‘speaker’ which is not explicitly marked within the sentence, cf. the example in (4).

In (4), Little Red Riding Hood believes that the one lying in the bed is her grandmother, whereas the reader knows that this is not in line with the reality of the story world. The denotation of the one lying in the bed as “the wolf” thus cannot reflect the perspective of Little Red Riding Hood, but an external viewpoint which has to be reconstructed from the communicative context. This is seen in the fact that the denotation as “the wolf” is not possible in the direct speech construction in (4b) that reflects Little Red Riding Hood’s perspective. (4a) thus integrates two different perspectives: the viewpoint of Little Red Riding Hood whose belief system does not contain the fact that the one lying in her grandmother’s bed is the wolf, and an external point of view which contains the knowledge about the fact that the one lying in bed is actually the wolf. The latter is not explicitly marked in the linguistic structure but can be reconstructed from ‘outside’.

At this point, it is important to be precise about the notion of internal vs. external. As applied to example (4), the distinction refers to the structural difference between two embedded viewpoints. In this sense, internal₁ captures the fact that the viewpoint is a second, displaced viewpoint that is viewed from outside by another external₁ viewpoint that is situated at a different level. This distinction is independent of the fact whether the viewpoint is linked to e.g. proprioceptive, kinaesthetic, and cognitive processes ‘inside’ one’s mind, i.e. to private feelings and thoughts that can be accessed directly via introspection (e.g. ‘I have a headache’) (i.e. internal₂). As will be seen in Section 4.3, both distinctions are relevant with respect to the perspectival structure but can lead to different evaluations of the viewpoint constellation.

In sum, it is thus trivially true that (1) is perspective-taking in the sense that a speaker uttering it chooses a certain perspective. But that is not the interesting point. Rather, the look on propositional attitude ascriptions shows that linguistic perspectivization is more than perspective-taking for more than one reason. First, as pointed out in the literature (Dancygier & Sweetser 2012; Dancygier & Vandelanotte 2016), even rather straightforward examples display not only one, but different viewpoints. Second, as also shown by Dancygier & Vandelanotte (2016), these viewpoints display hierarchical contrasts. With respect to the constellation of viewpoints, it is thus necessary to differentiate between equal alternatives that allow for perspective-taking in terms of an either or distinction (i.e. stage (ii) in Figure 2) and vertical relations of embedding (iii). Third, the integration of viewpoints requires a third external reference point from which the potentially contrasting viewpoints can be evaluated (iv). This calls for a three-point concept of perspective. As will be shown in the following section, the same is true for perspectivization in grammar.

In a wide sense of the term, grammar is genuinely perspectival in that every grammatical paradigm offers a choice of alternatives. In a more narrow sense, grammar is linked to perspectivization insofar as it locates the speaker’s deictic origo that determines the ‘view’ of the verbal event situation (cf. the approaches by Leiss 1992 and Diewald 2010 in tradition of Bühler and Jakobson). In cognitive linguistics, this is captured in the premise that “[i]nherent in every usage event is a presupposed viewing arrangement, pertaining to the relationship between the conceptualizers and the situation being viewed” (Langacker 2001: 16). It follows from this that grammatical markers do not indicate a certain viewpoint, but rather a viewpoint constellation (see Verhagen 2016; Dancygier & Sweetser 2012; Sweetser 2012; Dancygier & Vandelanotte 2016 for examples). This can be illustrated by a look at the category of tense. At first glance, tense markers like the simple past are “shifters” (Jakobson 1957) since they relocate the reference point from a present utterance time to the past. Hence, they could be seen as instances of perspective-taking. Moreover, such a displacement of the origo in the sense of “Deixis am Phantasma” (Bühler 1934 [1999]) relies on the persistence of the primary origo in the background. The temporal perspectivization thus refers to the relation between primary and secondary reference points and, hence, a multiperspectival constellation between viewpoints of different qualities. As such, it requires a three-point description, as reflected in the ternary system of Reichenbach 1947 that models tense as relations between the time of event and the time of speech as perspectivized by a third reference point (cf. in detail Zeman 2015; see also Evans 2005 with respect to the multiperspectivity of complex tenses).

Nevertheless, tense markers are prototypical shifters insofar as they shift the focus on the displaced viewpoint whereas the primary origo is only maintained in the background. This is different for markers of evidential and epistemic modality that display a more complex perspectival structure. As laid out in the introductory section, there seems to be a strong link between the development of ToM capabilities and the comprehension of epistemic and evidential modality that is a rather late development in language acquisition (Perner 1991: 150; Papafragou 1997: 16; Papafragou 2001; Leiss 2012). This becomes understandable by the fact that epistemic and evidential meanings share with propositional attitude ascriptions the feature that they are modifiers of propositions, and as such, overt markers of speakers’ attitudes. Both evidentiality and epistemicity thus display a similar perspectival structure by embedding veridical statements that may be true or false. This is seen in the fact that only (5b) but not (5a) can be translated into the structure of an epistemic modal verb.

In contrast to (5a), (5b) leads to the potential for viewpoint contrasts between the level of the sentence subject and the level of the speaker, and as such, the potential for an internal₁ and an external₁ view in the structural sense (i.e. inside vs. outside the proposition). This is also a crucial difference in comparison to the perspectival structure of root modals, cf. (6).

Root modals like in (6a) are characterized by their biphasicness, since they refer to two different time intervals, the temporal interval of the modal in the real world (i.e. the time for which the obligation holds) and the time interval of the event denoted by the infinitive complement in a possible world. As such, they display the potential for a focus shift from the present viewpoint to the future event, as seen in the fact that modals constitute a grammaticalization source for future tenses (cf. Zeman 2013). The main difference with respect to the epistemic example (6b) is yet that it introduces the speaker’s viewpoint, and, hence, an additional potential contrast between two belief systems (external₂ vs. internal₂): while the obligation in (6a) holds for the subject, i.e. Little Red Riding Hood, the modal meaning in (6b) scopes over the whole proposition p [Little Red Riding Hood is walking through the woods] and thus requires an external viewpoint of evaluation.

The perspectival complexity of (7b) is thus based both on the structural embedding of viewpoints (i.e. inside vs. outside the proposition), the distinction between two belief contents, and a contrast of veridicality on the propositional level, i.e. the fact that the proposition could be false. So once again, the construction integrates external and internal viewpoints at the same time. As such, they display a pattern of viewpoint integration (i.e. pattern (iv) in Table 1; Level 2b in the terminology by Moll’s & Meltzoff’s 2011). The difference in degree of complexity between (7a) and (7b) is also reflected in language acquisition. As laid out in Section 4.1, complementation structures like think that p are a later development than constructions like think of p that can be analysed in terms of perspective-taking. In analogy, verbs of volition are used prior to verbs of belief, and the usage of root modals precedes the comprehension of epistemic and evidential modals (cf. de Villiers 2005; Perner et al. 2005). In sum, it is not the potential viewpoint switch – and thus: not perspective-taking – that is crucial with respect to the difference between both constructions, but the viewpoint constellation that is based on the contrast between two different hierarchical levels (internal₁ vs. external₁), two different belief systems (internal₂ vs. external₂), and the contrast with respect to the veridicality attribution on the propositional level (true vs. false).

The degrees of perspectival complexity are also identifiable for perspectivization in narrative contexts. Narratives are perspectival structures par excellence since they integrate viewpoints of different characters and narrators and thus offer a set of possible alternate perspectives that allow for viewpoint switches within the text. However, also narrative perspectivization is more than perspective-taking for at least two reasons. First, the different viewpoints are not equivalent alternatives since the viewpoints of characters and narrators can be embedded in each other (cf. Dancygier 2012a; Dancygier & Vandelanotte 2016). This is in particular relevant with respect to the structural difference between narrator and character level since narration is characterized by the fact that the narrator level scopes over the event level of the protagonists (cf. also Zeman 2016; to appear).10 Second, these different viewpoints require in addition a global perspective from which the different perspectives are organized (see also Dancygier 2012; this volume; Dancygier & Vandelanotte 2016):

As a result, different layers of perspectivization arise. This can be best illustrated by a look at Free Indirect Discourse (FID), a perspectival pattern that is restricted to narrative discourse mode only and thus allows for some insights with respect to the structure of narration in general, cf. (8).

As (8) shows, FID blends together two different viewpoints (cf. Pascal 1977; see also Fludernik 1993: 316ff. for an overview; Vandelanotte 2009: 246–251 for discussion of ‘dual voice accounts’). While deictic elements such as tomorrow and the emotive choice of kinship terms like granny match the character’s perspective, personal pronouns and tenses are linked to the level of the narrator and thus allow for the reconstruction of the narrator’s viewpoint (cf. in detail Schlenker 2004; Sharvit 2008; Maier 2012; Eckardt 2014). The relationship between these two perspectives has been accounted for in terms of double context and is preferably analysed as (partial) context shift. In this respect, Eckardt (2014) has proposed to evaluate sentences in FID relative to a pair of contexts: “an external context C that is shared by the narrator and the reader, and an internal context c corresponding to a situation in which a protagonist of the story talks or thinks, as part of the story” (Eckardt 2014: 60).

As laid out before, the ‘internal’ vs. ‘external’ can refer to two different aspects, namely (i) the hierarchical difference between the communicative levels of discourse, whereby the narrator has naturally an ‘outside’ view on the character on the discourse level (whether this character is referentially himself or not) and (ii) the question of whether the contents on the propositional level are ‘thoughts’ that can be accessed directly (‘internally’). Due to its functional ‘outside’ position, the perspective of the narrator allows for simultaneous knowledge about his commitment towards the proposition and the course of the story which can include the mental contents of the protagonists. The narrator knows what the protagonist knows, while the reverse would lead to metaleptic structures. In a structural sense then, FID integrates both internal₁ and external₁ viewpoints at the same time.

On the other hand, however, FID is also characterized by the fact that on the discourse level, only one perspective is foregrounded, i.e. the character’s point of view, while the narrator’s voice remains ‘invisible’ and can only be traced back within the grammar, i.e. the use of tenses and pronouns.11 As a result, the reported propositional content in (8) is perceived as dependent on the character’s viewpoint – which could be reliable or not. So while FID structurally integrates two different perspectives (i.e. internal₁ vs. external₁), it is at the same time committed to the perspective of just one subject of consciousness that is restricted to mental contents of speech and thought (i.e. internal₂) (cf. Vandelanotte 2009: 246–251 for the argument that FID could be considered both as “bivocal” and “univocal”). As a result, grammatical constructions that are bound to the narrator’s perspective are incompatible with FID. This can be exemplified by the German ‘future of fate’ (FoF) construction that displays the complementary perspectival effect of FID:

While FID foregrounds the perspective of the character, the modal verb construction in (9) restricts the perspective to the narrator’s viewpoint. He knows what will happen next, whereas the character on the story level is unaware of the events to come. Such a view from outside is incompatible with the character perspective which seems to be the cause why the FoF-reading is blocked in FID (cf. Eckardt 2012: 185). In a sentence like She was glad. Tomorrow she should see her again! the event to come is focalized from the character’s perspective, linked to a less degree of certainty than the proleptic prediction by the narrator. The hierarchy of perspectives is seen in the fact that the character is subordinate to the force of the narrator’s point of view whose controlling authority is grounded on its functional role within the narrative discourse. The authority of the narrator makes the event realization highly certain and rules out any intervention of the character on the story level. Hence, both constructions – FID and FoF – act on the double context of narration and structurally integrate a potential contrast of viewpoints, but the viewpoint constellation is reverse in that one perspective is suppressed in favor of the other on the discourse level.

So is this perspective-taking? Yes and no. Apart from the trivial fact that it is perspective-taking on the side of the speaker who selects a certain perspective in choosing either FID or FoF (or any other option), one could argue that it is perspective-taking in the sense that the perspective either of the narrator or the character is ‘taken’. On the other hand, it is more than just perspective-taking, since both (8) and (9) refer to more than one subject of consciousness (i.e. internal₂ PoVs that are linked to mental agents) and rely on a hierarchical contrast between external₁ and internal₁ PoVs (in the structural sense of the term).

Furthermore, both constructions are based on a contrast between the narrator’s and the character’s viewpoint. In FoF, the divergent states of knowledge trigger a narrative tension since the narrator (and the reader) knows more than the character about the things to come. In FID, the report from the character’s perspective could turn out as unreliable or create ironic effects in contrast to the narrator’s viewpoint. In both cases, the potential contrasts thus call for a third evaluating viewpoint on a global level from which the perspectives are monitored. This is in line with Klein & von Stutterheim (2002) who distinguish between local perspectives within the text and “a globally established perspective” (Klein & von Stutterheim 2002: 83) that maintains the quaestio of the story. With respect to narrative discourse structure in particular, Dancygier (2012a), this volume, has argued that narratives are structured in different narrative spaces that require “a high-level story-viewpoint (SV) space, which structures the text-wide viewpoint of the story” (Dancygier 2012a: 129f.). That is to say, a global viewpoint is required that monitors the different perspectives in order to establish a coherent view on the story.

The establishment of such a global viewpoint is also a fundamental developmental step, as it is well known that stories of young children commonly lack a coherent macro-organization (cf. von Stutterheim & Carroll 2007: Fn. 3, with reference to Berman & Slobin 1994; see also Büttner 2016 with respect to disorders of macro-structural planning from a neurolinguistic perspective). This does not only affect the production, but also the reception of stories. According to Goldie (2007), the capability of taking an “external perspective”, i.e. thinking about another person in terms of “‘he thinks that p’ and ‘he thinks that if p then q’” (Goldie 2007: 70), constitutes the prerequisite for the appreciation and evaluation of diverging perspectives in narratives (cf. similarly Zunshine 2006 with respect to the role of ToM-task in narrative appreciation). Thereby, the comprehension of ‘dramatic irony’ displays more than a shift of perspective since the reader has to take into account (at least) two diverging viewpoints simultaneously in order to appreciate the story: the reader has to understand that the one in bed that Little Red Riding Hood is seeing is actually the wolf while knowing at the same time that Little Red Riding Hood thinks that it is her grandmother. Dramatic irony thus displays the same viewpoint constellation as laid out in Section 3 for false belief understanding: The beliefs of two mental subjects as alternate viewpoints that have to be evaluated from a third reference point as contrasting ones with respect to their propositional content and evaluated with respect to their relation to each other.