Experiential Imagination and the Inside/Outside-Distinction

Abstract

Gerundive imagination reports with an embedded reflexive subject (e.g. Zeno imagines himself swimming) are ambiguous between an ‘inside’ and an ‘outside’ reading: the inside reading captures the imaginer’s directly making the described experience (here: swimming); the outside reading captures the imaginer’s having an experience of an event, involving his own counterpart, from an out-of-body point of view (watching one’s counterpart swim). Our paper explains the inside/outside-ambiguity through the observation (i) that imagining can referentially target different phenomenal experiences – esp. proprioception (i.e. bodily feeling) and visual perception (seeing, watching) – and (ii) that imagining and its associated experience can both be de se. Inside/outside readings then arise from intuitive constraints in the lexical semantics of verbs like feel, see.

We thank two reviewers for LENLS 17 for valuable comments on the précis of this paper. The paper has profited from discussions with Alex Grzankowski and Ede Zimmermann. Kristina Liefke’s contribution was supported by the German Research Foundation (DFG) through grant no. ZI 683/13-1 (to Ede Zimmermann). Markus Werning’s contribution is supported by DFG grants no. 419038924 and no. 419040015 as part of the DFG research group Constructing Scenarios of the Past (FOR 2812).

1 Introduction

Imagination reports like (1) are generally taken to have two different kinds of de se-reading (see e.g. [2, 39,40,41]): an inside (subjective, or experiential) reading,¹ which captures what it would be like for the imaginer to undergo the described experience; and an outside (objective, or imagistic) reading, which captures what it would be like for the imaginer to witness an event, that involves his own counterpart, from an out-of-body point of view. The inside reading of (1), i.e. (1a), reports a relation towards the bodily experiences of Zeno’s swimming counterpart (e.g. the salty taste of the water, the tug of the current, the feeling of cold). The outside reading, (1b), reports a relation towards the target of Zeno’s (counterfactual) visual perception that has Zeno’s swimming counterpart as its object (e.g. an observer view of Zeno being tossed about, his body bobbing up and down in the foamy waste; [39, p. 161]).

Recent work on self-imagination reports (esp. Anand [2] and Ninan [24]) explains the ambiguity in (1) through the de dicto/de se-distinction [24] (see [13]) or through the particular way in which we set up imaginative projects [2] (see [46]). However, this work either fails to capture the experiential character of imagining (in Anand’s case) or the perceptival nature of the outside reading (in Ninan’s case). Specifically, Ninan’s account (dubbed the Simple View in [24]) counterintuitively treats the outside reading of (1) as equivalent to (2):

Our paper seeks to compensate for the above shortcoming. To do so, it uses a variant of Blumberg’s [5] observation (see also [25, 34, 40]) that imagining can be referentially dependent – or parasitic – on experiences. Our variant involves the inverse of Blumberg’s referential dependency relation, viz. the dependency of some counterfactual experiences (e.g. counterfactual proprioception or visual perception) on imagining. The ambiguity in (1) can then be explained through the fact that imagining and its dependent experience(s) can both be ascribed de se. It arises from the existence of intuitive constraints in the lexical semantics of proprioception and perception verbs.

The paper is structured as follows: we start (in Sect. 2.2, 3.1) by describing two properties of imagining that are particularly relevant for the inside/outside-ambiguity, viz. experiential parasitism and de se-ness, and argue that these properties can be co-instantiated in a single imagination report. We then show that the formal-semantic tools that are commonly used to capture these properties, viz. world-variables in syntax (see [27]) and centered worlds [13], can be straightforwardly combined into a single formalism (Sect. 3.2). Section 4 uses the content- and act-specific properties of proprioception and perception to reduce the many possible LFs of (1) that our formalism predicts to the inside- and the outside reading. Section 5 identifies the grounds for Vendler and Walton’s disagreement about outside readings of gerundive imagination reports with a PRO subject and explains the non-availability of an outside reading of Williams’ [46] imagine being Napoleon. The paper closes by pointing out that the linguistic inside/outside-distinction may reflect a real psychological and neurobiological difference.

2 Experiential and Parasitic Imagining

Before we discuss the properties of imagining that are relevant for the inside/out-side-distinction, it is important to identify the particular kind of imagining that this paper is about:

2.1 Experientiality

We have suggested above that our considerations in this paper focus on experiential imagining (esp. imagining feeling/seeing). The latter is an event-directed attitude – similar to experiential [= episodic] remembering (see [36, 37, 43]) – that requires the attitudinal agent’s personal (counterfactual) experience of the target event or scene (see [8, 34]). For the memory report in (3a) and a variant, (4a), of the imagination report in (1), the satisfaction of this requirement is evidenced by the validity of the inferences in (3) and (4):

As is suggested by our use of the verb experience in (4b), we assume that experience does not entail or presuppose the truth of its complement (i.e. experience is neither veridical nor factive; see [9]). To still explain the factivity of remember in (3a) (attested, e.g., in [12, 45]), we observe that remember is derivative on the particular mode of the remembered experience (e.g. visual [see (3b)], proprioceptive, agentive, emotional; see Sect. 2.2). The relevant occurrence of remember then inherits the veridicality and factivity properties of this mode.

Importantly, in contrast to episodic remembering, experiential imagining can also be reported through that-clause complements (see [15, 23]; pace [34]).² This is reflected in the fact that that-clause-taking imagine allows modification with event-modifiers like vividly (see (5b) and the data in [38]). It has been argued that such modification is not possible for that-clause memory reports (see (6b)) and that those are associated with propositional, i.e. semantic, memory (see [34]).

The above suggests that the inadequacy of Ninan’s paraphrase in (2) (which uses a that-clause) is not due to the syntactic form of its complement, but to the particular inserted predicate (i.e. doing/experiencing). We will return to this point in Sect. 4.

2.2 Counterfactual Parasitism

Recently, Blumberg [5] and Ninan [25] have argued that experiential imagining can be parasitic [= referentially dependent] on experiences, in the sense that the ‘correct’ analysis of experiential imagining requires some imagination contents to take their referents from worlds other than the actual/evaluation world or the agent’s imagination alternatives.³ This analysis is prompted either by the presence of experience predicates (in (7) and (8): see resp. dream) or by the lack of adequate truth-conditions in the absence of these predicates (thus, (9a) requires the insertion of a dream-PP; in (9b)). Examples (7) and (8) are due to Ninan [25, ex. (18)] and Blumberg [5, ex. (102)], respectively. Example (9) is inspired by Blumberg’s [4] ‘burgled Bill’-case. In what follows, we mark the matrix [= ‘parasite’] attitude (here: imagining) with a grey frame. The experience [= ‘host’] (i.e. seeing resp. dreaming) is highlighted in grey:

The parasitic interpretation of imagine in (9a) is triggered by the observation that – given the context in (9) – (9a) is false on its de re-reading (which gives the DP her [= a tattooed woman] a specific interpretation; see (9a-i), (10a)) and that (9a) is contradictory on its de dicto-reading (see (9a-ii), (10b)). The parasitic interpretation is then prompted by the observation that (9a) has plausible truth-conditions on a reading that evaluates her at some other world (different from the actual world and from Ira’s imagination alternatives; see (9b), (10c)). Our name for this reading, i.e. de hospite, is motivated by the observation that this world is associated with the host experience, on which the matrix attitude (here: Ira’s imagining) is parasitic. In (10c), the ‘host’ world is denoted by X:⁴

To specify the particular world(s) at which the different constituents of the complement in (9a) are evaluated, (10) uses Percus’ [27] Index Variables-approach. This approach posits possible world-variables in the representation of syntactic structures, and allows intensional (here: attitude/experience) operators to bind these variables. In particular, Percus’ approach assumes that all predicates contain an unpronounced variable that saturates their world-argument. It further assumes that intensional operators are associated with a lambda abstractor that can bind a world variable. The ability of the same world variable in a syntactic structure to be bound by different lambdas then accounts for different readings.

The LFs in (10) only assume a single world-variable, w, next to our variable for the actual world, @. To capture our observation that the constituents of the complement in (9a) are dependent on different worlds/alternatives, we follow Blumberg [4] in positing distinct variables for the alternatives that are introduced by the parasite attitude [here: imagining] ($w_{2}$) and for the alternatives that are introduced by the host experience [here: dreaming] ($w_{1}$). The different readings of the imagination report in (9a) (see (10)) are then associated with the LFs in (11). The relevant LF – on which (9a) is true – is given in (11c).

We have suggested above that, in (11c), the matrix attitude [= imagining] depends for its reference on the underlying experience [= dreaming]. The direction of this dependence motivates the ‘parasite’/‘host’-terminology in [5] (see [4], due to [16]). The situation is different for imagination reports like (1): arguably, in such reports, the embedded subject DP (in (1): himself) is still interpreted at the imaginer’s experience alternatives (viz. at Zeno’s proprioception- or perception alternatives). However, in these reports, the referential dependency is the other way around, i.e. the implicit experience is dependent on the imagining.⁵ In particular – unlike (9) –, (1) does not assume that the imaginer’s experience [there: Zeno’s feeling or seeing] happens in the same world as his imagining (viz. at @). Rather, it only happens in his imagination. To capture the inverse dependency relation of reports like (1) w.r.t. de hospite-reports, we describe reports like (1) as de parasito. The inverse dependency of the matrix attitude and the experience in (1) validates the equivalence in (12) (see [40]), where ‘V’ stands proxy for the experience (i.e. $V \in \{\textit{feel}, \textit{see}\}$):

The above suggests that (1) should not be analyzed as an analogue of (11c) (i.e. as (13)), but rather as (14a). This LF inverses the order of the lambda abstractors over imagination- and experience alternatives (in comparison to the order of the lambda abstractors over imagination- and dream alternatives in (11c)).

Notably, in contrast to (11c), the LF in (14a) interprets the embedded predicate swim at the same world as the embedded subject DP himself, viz. at the imaginer’s experience alternatives, $w_{1}$. This is required by the assumption that the described event (here: a swimming) takes place in the same counterfactual world at which the agent of this event (here: the referent of the embedded subject DP; i.e. Zeno’s counterpart) is determined. To capture the equivalence in (12), (14b) makes explicit reference to the dependent experience (see the clause ‘t Vs in $w_{2}$’). Since this reference also requires identifying the subject of this experience (here: Zeno[’s counterpart], or – as we will see later – the semantic value of the silent pronoun PRO), it facilitates the formal implementation of first person-perspective (see Sect. 3.2). In (14b), the LF-referents of the ‘imaginer’ and the ‘experiencer’ are circled in grey (imaginer) resp. in black (experiencer).

We close this subsection with a remark on the compatibility of (14) with Percus’ Generalization X. The latter is a constraint on admissible readings of a sentence which demands that the world variable that a verb selects for must be coindexed with the nearest lambda above it [27, p. 201]. This constraint excludes (15a) as an admissible reading of (1). This LF gives a reading that describes Zeno’s experience-counterpart as swimming in the actual world, @:

Since the mere inversion of the order of the lambdas in (13) (see (15b)) evaluates swim at the ‘middle’ world, $w_{2}$, Generalization X also excludes the reading in (15b). By interpreting the embedded predicate at the same world as the embedded subject DP (along the lines proposed in (14)), we avoid this exclusion. In (14), swim is evaluated at the ‘lowest’ world, $w_{1}$, as Generalization X demands.

3 Self-Imagination and Experiential Parasitism

With the referential dependence between imagination and experience(s) in place, we turn to the second property of imagining that is relevant for the inside/outside-distinction, viz. de se-ness:

3.1 Imagining de se

De se- (or self-locating) attitudes are first-personal attitudes that the holders of these attitudes self-ascribe, to the effect that these attitudes “crucially involve the attitude holder’s access to [his/her] own ‘self’” ([33, p. 411]; see also [13, 34]). In English, de se-attitudes are commonly denoted by reports with subject-controlled infinitives or gerundives (e.g. (16b), (17); see [6]) and can be denoted by infinitives and gerundive small clauses with a reflexive subject (e.g. (16); see [28]):

Following Lewis [13] and Chierchia [6], the contents of de se-attitudes are standardly modelled as sets of centered worlds. Centered worlds are worlds that are experienced from the perspective of one of the individuals in these worlds (i.e. from the perspective of the center of these worlds). Formally, centered worlds are coded as world/individual-pairs $\langle w, y\rangle$, where y is the center of w (see [13, 33]).

In attitude reports like (16b) and (17), Chierchia’s analysis associates the subject of the control clause, i.e. PRO, with the individual center of the world that is introduced by the matrix attitude verb (above: want resp. avoid; see [6]). Analogously to the treatment of world-variables in syntax (see Sect. 2.2), variables over individual centers can be bound by a lambda abstractor. To emphasize the ‘unity’ of centered worlds, we allow abstraction over ordered pairs of world- and individual-variables, resulting in abstracts of the form $\lambda \langle w, y\rangle$. Using such abstracts, the reports in (16b) and (17) are then analyzed as (18) and (19), respectively:

Unsurprisingly, the above analysis is often also applied to imagination reports with subject-controlled gerundive small clauses (e.g. (20a); see [10, 34]). On this analysis, the ‘non-parasitic’ reading of (20a) (which neglects the dependent experience) is taken to report Zeno as standing in the imagining relation to worlds whose center is swimming (see (20b)):

3.2 Experientially Parasitic de se-Imagining

The analysis of the subject-controlled imagination report in (20b) can be straightforwardly transferred to the de parasito-version of (20a), i.e. (21a). In the resulting LF (see (21b)), the de se-center is doubly circled. To avoid overly long LFs, we replace ‘t’s counterpart-in-$w_{1}$’ by ‘t-in-$w_{1}$’:

Note that the LFs in (20b) and (21b) identify different centered contents. In particular, while (20b) interprets the complement in (20a) as the set of centered worlds whose center is swimming, (21b) interprets this complement (analyzed as the imagine-complement in (21a)) as the set of worlds whose center is V’ing (e.g. visually perceiving) Zeno’s counterpart from the imagined scene swimming. We will return to this difference below.

Work on self-imagining (e.g. the kind of imagining reported by (1)) typically follows the above in identifying the de se-attitude with the matrix attitude (see e.g. [34, 40]). What has escaped researchers’ attention – but what is at work in (1a) vis-à-vis (1b) – is that the experience can also be de se. This is suggested by our discussion of parasitic imagining in Section 2.2 and is evidenced by (22) (note the silent pronoun PRO in the complement of V):

The possible de se-ness of the matrix attitude [= imagining] and the experience [= V’ing] then predicts four combinatorially possible parametrized LFs for (1) (in (23), where (23a) copies (14b) and where $V = \{\textit{feel}, \textit{see}\}$). Since we assume that names are rigid designators (s.t. evaluating Zeno yields the same individual at all worlds), we suppress counterpart relations (writing ‘t’ instead of ‘t-in-$w_{1}$’).

In what follows, we call the above LFs ‘non-de se’ (i.e. (23a)), ‘experience de se’ (i.e. (23b)), ‘matrix de se’ (i.e. (23c)), and ‘doubly de se’ (i.e. (23d)), respectively. The LFs in (23) roughly correspond to the English sentences in (24):

4 Multiply Parasitic Imagining and Constraints on de se-Ascription

We have suggested above that imagining can determine different counterfactual experiences – saliently, proprioception [bodily feeling] (see (1a)) and visual perception [seeing/watching] (see (1b)). Given the possible realization of V by feel respectively by see, the parametrized LFs in (23) then have the (many !) possible readings in (25):

A first restriction on the readings in (25) comes from natural constraints on the content of proprioception resp. of visual perception. These constraints include the obligatory de se-nature of proprioceptive content (i.e. the inherently first-personal perspective – or self-directedness – of bodily feeling) and the typically non-de se nature of visual perception content (i.e. the observation that our vision is typically directed towards the outside). The first constraint excludes all LFs with non-centered ‘feeling’-content, viz. (25a) and (25c) (indicated by a double strikethrough). The second constraint marks as non-salient all LFs with centered visual perception content, viz. (25f) and (25h). Vendler attributes this non-salience to the external perspectivity of visual perception, which “put[s] the perceiver in a spatial relation to the object” [39, p. 165]. In (25), LFs with centered visual perception content are indicated by a single strikethrough.

From the remaining LFs (copied in (26)), the intuitive readings of (1), i.e. (26d) $[=$ (1a)] and (26b) $[=$ (1b)], are then obtained by considering intuitive lexical-semantic constraints on acts of proprioception respectively of (visual) perception. These constraints include the inherently de se-nature of bodily feeling and of seeing. This nature excludes non-self-locating feeling and seeing, as is assumed in (26c) and in (26a) (indicated by a single strikethrough):

The combination of ‘V $=$ feel’ with centered matrix and experience content – and the attendant identification of the ‘experiencer’- with the ‘swimmer’-perspective in (26d) – then identifies (26d) with the inside reading of (1). The combination of ‘V $=$ see’ only with centered matrix content – and the attendant separation of the (centered) ‘perceiver’- and the ‘swimmer’-perspective – identifies (26b) with the outside reading.

5 Applications

We finish our paper by using the proposed analysis to account for some well-known puzzles and debates involving the inside/outside-distinction. These include Vendler and Walton’s disagreement about outside readings of gerundive imagination reports with PRO-subjects (see Sect. 5.1), the salience of inside readings of subject-controlled gerundive memory reports, and the non-availability of an outside reading of Williams’ [46] imagine being Napoleon (both Sect. 5.2):

5.1 Vendler and Walton’s Disagreement

Our previous considerations have focused on imagination reports with reflexive complements (i.e. complements that are headed by reflexive pronouns like himself). While most researchers agree that such reports are ambiguous along the lines described in Sect. 1, they disagree whether imagination reports with subject-controlled gerundive complements (e.g. (20a), copied in (27)) display an analogous ambiguity: in line with Vendler [39, pp. 162–163], many researchers assume that subject-controlled gerundive imagination reports only allow for an inside reading (see e.g. [26, 34]), making (27) equivalent with (27a). Following Walton [41, pp. 28–35], researchers in the opposing camp assume that (27) can also be used to report an outside perspective (see e.g. [2, 44]), making (27) ambiguous between (27a) and (27b) (with a slight preference for (27a)):

The acceptance of outside readings like (27b) is typically fuddered by examples like (28), which include the perspective of the spectator (in (28a), due to Walton [41, p. 31]) or which remove the possibility of inside consciousness (in (28b); due to Anand [2, p. 4]; see [39, p. 166]):

However, upon closer inspection, the examples in (28) do not support the availability of (27b) as an admissible reading for (27). This is due to the fact that, in the scenes that are described by these examples, the familiar interpretation of subject-controlled gerundive complements (= ‘inside’) is not available. The outside reading of (28a) and (28b) can then be explained through a pragmatic reinterpretation of the complement in these reports. Since (27) – by admission of defenders of Walton’s ambiguity – has an inside reading, an analogous explanation of the outside reading is not available for (27).

The above suggests that Walton’s predicted ambiguity is due to a flip-flop between a syntactically suggested inside reading and a pragmatically coerced outside reading. Our results from Sects. 2, 3 and 4 suggest an alternative explanation of Vendler and Walton’s disagreement about the ambiguity of (27) that does not assume pragmatic coercion. This explanation is based on the existence of two possible referents of PRO in (27), viz. the individual center of the alternatives that are introduced by the matrix verb (s.t. (27) is analyzed as ‘matrix de se’; see (29a)) or the individual center of the alternatives that are introduced by the silent experience verb V (s.t. (27) is analyzed as ‘experience de se’; see (29b)):

The readings in (29a) and (29b) differ with regard to which LFs they allow for (29): while the matrix de se-reading in (29a) – which we associate with Walton – is compatible with both of the LFs in (30) (see (23c), (23d)), the experience de se-reading in (29b) – which we associate with Vendler – is only compatible with the LF in (30b) (see (23d)). The unavailability of this LF for ‘V $=$ see’ (see our argument for the exclusion of (25h) in Sect. 4) then explains Vendler’s exclusion of an outside reading of (27). The availability of (30a) for ‘V = see’ explains Walton’s inclusion of this reading.

Arguably, the above observations still leave open the question of which one – matrix or experience de se – is the ‘correct’ reading of PRO in (27), i.e. who was right: Vendler or Walton. Since the syntax of iterated attitude reports is still understudied (s.t. we cannot draw any conclusions about matrix or experience de se based on the movement and ellipsis behavior of (24)), we try to answer this question by considering a closely related domain, viz. gerundive remember-reports:

5.2 Remembering ‘From the Outside’

In contrast to gerundive imagination reports (e.g. (20a), (1)), gerundive memory reports are generally taken to have a salient inside reading (see (32a)):

The salience of the reading in (32a) is supported by the observation – reflected in corpus data – that memory reports with subject-controlled gerundive complements (e.g. (31)) are strongly preferred⁶ over memory reports with reflexive subject-complements (e.g. (32)) in most contexts. This observation is striking since the pronoun himself is typically ambiguous between a control and a non-control interpretation (see [28]). As a result, one would expect that the admissible readings of (32) include⁷ the reading(s) of (31). Given speakers’ general dispreference for (32), this suggests that remember semantically marks (as deviant, or ‘noteworthy’) non-centered alternatives in either matrix or experience position.

We will see below that our constraints on the content and act of the experience (i.e. $V \in \{\textit{feel}, \textit{see}\}$; see Sect. 4) attribute this marking to non-centered alternatives in experience position. We have already found that these constraints only leave the imagine-counterparts of (see (26d)) and (see (26b)):

Since, in (33), the only LF that is not doubly de se (i.e. (33b)) has its non-centered alternative in experience position, we conclude that remember marks non-centered alternatives in this position. This suggests that the explicit PRO in (31) is interpreted in experience position. This is in line with Vendler [40] and with the intuition of most researchers on the inside/outside-distinction (see Sect. 5.1).

We finish this section by suggesting an explanation for the non-availability of an outside reading of Williams’ imagine being Napoleon: in [46, p. 43], Bernard Williams observes that, while he can imagine from the inside being Napoleon (s.t. he can hold the attitude that is reported by the reading of (34) in (34a)), he is unable to imagine from the outside that he is Napoleon (s.t. he cannot hold the attitude that is reported by the reading in (34b)). Williams backs his observation by referring to the intuitive absence of a self – distinct from (himself qua) Napoleon – that could perceive this identity.⁸

Our framework captures Williams’ intuition through another familiar constraint on visual perception: the restriction to what can be visually perceived. Specifically, this constraint includes that perception cannot serve to establish the personal identity of a perceived object (e.g. Napoleon) with the perceiver. However, exactly this would be required for the outside reading of (34). The insufficiency of perception for the establishment of personal identity is reflected in the semantic deviance of (34b).

The above notwithstanding, our identification of PRO in (34) with experience de se (see above) even allows for a yet simpler exclusion of (34b). The latter is based on the fact that (34b) involves experience non-de se (see the overt [= non-controlled] occurrence of he in the complement of see). Since this is incompatible with the use of PRO in (34) – as we have argued for (31) –, (34b) is not an admissible reading of (34).

6 Outlook

Our considerations in this paper have focused on the linguistic realization of the inside/outside-distinction. The cross-linguistic robustness⁹ of this distinction suggests that there is a real psychological difference between the first-personal (field) and the third-personal (observer) perspective on a personally experienced event. For imagination, this difference is already suggested in Vendler [39], and has been corroborated by behavioral and imaging studies (see e.g. [7, 14]). In particular, in [7], Christian et al. have shown that first-person imagining of painful scenarios elicits greater activity in brain areas associated with interoceptive and emotional awareness, with visual imagery, and with sense of body ownership.

In their groundbreaking work, Addis et al. [1] and St. Jacques et al. [32] have observed a remarkable overlap in the neural and cognitive mechanisms that underlie episodic memory and imagination. Michaelian [21] even goes so far as to claim that remembering is just a special form of imagining that results from a reliable episodic construction mechanism and is directed towards one’s personal past. Pointing out the close connection between the reliability of remembering and its causal dependence on the event remembering, Werning [42] contradicts this view and argues that remembering is distinct in kind from imagining.

Regardless of this controversy, the overlap between neural and cognitive processes underlying remembering and imagining might suggest that the difference between first- and third-person perspective on the experienced event in imagination is equally present in episodic remembering. This suggestion is further supported by the fact that the verbs remember and imagine have a very similar selection behavior. However, at least for recent events, observer perspective has been found to be less common in episodic remembering [20, 31] (see [29, 30]). This may be due to the particular importance of self-performed actions for episodic memory, and to the inherently first-personal perspective on such actions. Exceptions to this rule are PTSD patients’ memories of traumatic events [3] and memories that involve intense emotional components or high self-awareness [22, 30].

Recently, McCarroll [17, 18] has claimed that agents can also take a third-personal mnemonic perspective on non-traumatic and emotionally less intense events. McCarroll supports his claim with reference to the epistemic generativity of episodic memory and to the observation that observer memories can be epistemically and emotionally beneficial. We leave the exploration of this claim as a topic for future work.