R1.1. The nature of dreaming

I argue the nature of dreaming results from its highly associative content.

Hobson describes my hypothesis as neo-Freudian because it puts personally meaningful association (sand–house–bridge) at the centre of dream construction. Association condenses episodic memories. Erdelyi, although drawing on Freudian concepts of fusion and condensation, describes these associations as memory errors only, disregarding cited evidence on the centrality of association for memory, although the highly associative nature of memory (Vutter–Mutter–Vater) does, indeed, sometimes lead to retrieval errors. Porte also relates my dream hypothesis to that of Freud, seeing them both as “high,” “top-down” theories (para. 2). I argued that REM dreaming, like waking, is self-organized. There is no central “top-down” generator. Porte suggests, instead of “high” theory, a puzzling “egalitarian” explanation grounded in cholinergic activation. As my theory is about the function of REM dreaming, even while her exposition of the cholinergic system and its role in dreaming is useful and accurate, I don't comprehend how cholinergic activation explains the content, as opposed to the form, of dreams.

Another puzzle is Solms's assertion that I take the position of REM/dream isomorphism. To clarify, I argue that REM and NREM dreams are characterized by unique physiology and dream structure and, for many purposes (i.e., not all purposes; see sect. R5) should be considered separately. Solms (Reference Solms, Pace-Schott, Solms, Blagrove and Harnad2000/2003b) also differentiates REM and NREM dreams, commenting, “Few people would disagree that the average NREM dream is more ‘thoughtlike’ than the average REM dream” (p. 247). On the other hand, as I acknowledge, it is hard to distinguish late-night NREM from early-night REM dreams. My hypothesis points to an explanation of this seeming paradox. I elaborate in section R5.

Greenberg (see also Erdelyi) argues that the target article is incomplete because, although both the neurobiology and psychology of dreaming are reviewed, there is no thorough review of the clinical research on dreams and dreaming – without this, one leg of Greenberg's three-legged stool is missing. But, as Cicogna & Occhionero point out, empirical research on dreams and dreaming is scant and inconsistent – the third leg is distinctly wobbly. This is one of the reasons I illustrated with “Quicksand” rather than cover the clinical research on dreaming.

Erdelyi lists several analogies that can be made with dreams and dreaming: aphasia, subliminal perception, hysterical symptoms, schizophrenia, jokes, daydreams, poetry, and Bartlettian and Freudian reconstruction. I agree. I cannot comment on all of these, but I suggest that, rather than disparate to each other and memory, they are all driven by association (or, in the case of aphasia and schizophrenia, disruption of associative processes). Association is fundamental to all forms of memory (e.g., see Fuster Reference Fuster1999).

R2.1. Dreaming and creativity

I agree with DeYoung & Grazioplene that, first, emotional associativity is not silenced in wake but merely subordinated, and, second, this suppression is less in more creative individuals (such as writers) who may be elaboratively encoding their memories in wake (cf. Cercas Reference Cercas2001/2004). This is a testable proposition because, as DeYoung & Grazioplene point out, creativity in wake is probably enabled by the waking neuromodulatory balance shifting toward dreaming. But DeYoung & Grazioplene do not explain why this shift occurs, although they note individual variation in all neuromodulators and, in DeYoung et al. (Reference DeYoung, Grazioplene and Peterson2012), the increased risk for psychosis in creative individuals.

In Llewellyn (Reference Llewellyn2011), I say why shifts in neuromodulatory systems occur and how these changes could engender what I term “de-differentiation” between waking and dreaming consciousness. I argue: “Within every 24 h, two self-organized, highly activated states are achieved: waking and dreaming. Therefore, in the mind/brain, it is intuitively likely that ‘order’ is represented by an achieved differentiation between waking and dreaming and ‘chaos’ (or disorder) results from their de-differentiation” (p. 1062). De-differentiation is a matter of degree, however. When the mind/brain is poised at self-organized criticality between order and disorder, it is optimized for complex tasks (Kauffman Reference Kauffman1993, p. xv). In a state of “super-criticality” (see Pearlmutter & Houghton, para. 3) between waking and dreaming, the waking mind may be able to make the creative, imaginative leaps of association that, usually, are possible only during dreaming. This would foster creativity, which sees non-obvious associations (i.e., similarity between, seemingly dissimilar, things). Self-organized criticality implies that even small system perturbations can have large, unpredictable, and, sometimes, deleterious effects (Bak Reference Bak1996; Lewin Reference Lewin1993, p. 11; Orsucci Reference Orsucci2006). This may explain the link between creativity and psychopathology. In more creative individuals (who are in a state of super-criticality), further, albeit small, increases in the neuromodulators that characterize REM dreaming (dopamine and acetylcholine) may increase de-differentiation and precipitate psychopathology. In consequence, small differences in initial neuromodulatory conditions can engender very different outcomes: creativity or psychopathology.

R2.2. Dreaming and psychopathology in wake

D'Agostino & Scarone note that similarities between dreaming and psychosis have been remarked upon since Aristotle, and argue that dreams are the stuff that psychoses are made on. In Llewellyn (Reference Llewellyn2011) I agree – the intrusion of the dreaming state into wake may precipitate psychosis. Psychotic individuals may also be encoding their memories in wake through hyperassociation. Lake (Reference Lake2008) demonstrated that careful observation and questioning of a patient exhibiting “loose associations,” “derailment,” and “flight of ideas” revealed the chain of personal associations that were driving his thinking. The difference between psychopathology and creativity may be one of degree. In psychosis, dreams are uncritically accepted as real, probably because REM dream physiology, including deactivation of the critical judgement and reality monitoring functions of the dorsolateral prefrontal cortex, characterize the waking state. Solms asks why we usually stay sane when remembering dreams in wake, simply because critical judgement and reality monitoring are switched on, so dreams are recognized as dreams.

I argue that D'Agostino & Scarone are considering only one side of a de-differentiation equation. In Llewellyn (Reference Llewellyn2011), I review research on dynamic reciprocity among neuromodulatory systems, concluding that de-differentiation between waking and dreaming increases mesolimbic dopaminergic and cholinergic neuromodulation in wake, which may precipitate psychosis, whilst, simultaneously, decreasing these neuromodulators in dreaming. The latter may impede encoding in dreams. D'Agostino & Scarone point to Laviolette (Reference Laviolette2007), who argues that dopaminergic transmission encodes the emotional salience of events and is more ubiquitous in memory processes than has been recognized previously. The mnemonic role of acetylcholine is well established. Specifically, drugs that block acetylcholine receptors impair encoding and cholinergic effects have been argued to be important for the hippocampal encoding of episodic memories (e.g., see Hasselmo Reference Hasselmo2006). If REM dreaming elaboratively encodes episodic memories, decreases in cholinergic input and mesolimbic dopamine, consequent upon the de-differentiation of waking and dreaming, would impair memory hyperassociation.

Episodic memory enables both the retrieval of emotionally salient events from the past and the capacity to envisage the future (Schacter et al. Reference Schacter, Addis and Buckner2007; see also Girard). Indeed, the core attribute of episodic memory may be the ability to vividly imagine and experience the agential self in time (Szpunar et al. Reference Szpunar, Watson and McDermott2007; Tulving Reference Tulving1993). Markowitsch & Staniloiu rightly point out that my definition of episodic memory should have highlighted this re-experiencing. Episodic envisaging would underpin the sense of self-continuity over time, a commitment to an emotionally salient to-be-achieved programme of action and the initiation of purposeful action in relation to this. Memory impairments in schizophrenia are episodic rather than semantic (Huron et al. Reference Huron, Danion, Giacomoni, Grangé, Robert and Rizzo1995; Tendolkar et al. Reference Tendolkar, Ruhrmann, Brockhaus, Pukrop and Klosterkotter2002). The negative symptoms of schizophrenia have been conceptualized as apathy, social isolation, diminished affect, and lack of motivation (e.g., see Andreasen Reference Andreasen1990). I noted Solms and Turnbull's (Reference Solms and Turnbull2002) observation that patients who do not dream become “aspontaneous, inert and apathetic” (p. 312). Thus, there is similarity between the symptoms of non-dreaming patients and the negative aspects of schizophrenia. This suggests that the de-differentiation of waking and dreaming consciousness, resulting in loss of functionality for waking and dreaming, may engender not only positive (psychotic) symptoms in wake but negative symptoms also and, hence, ultimately, schizophrenia.

R2.3. Psychopathology and sleep disorders

Two commentaries discussed psychopathology and sleep disorders. Van der Kloet, Merckelbach, & Lynn (van der Kloet et al.) are interested in whether excessive REM is related to dissociative symptomatology. Girard focuses on the specific nature of hallucinations in sleep paralysis. I concur with van der Kloet et al. that dissociative symptoms are probably engendered by a labile sleep–wake cycle where dream-like mentation invades the waking state. Like D'Agostino & Scarone, I argue that, in psychosis, this dream invasion into wake occurs in a more severe and enduring form. I think van der Kloet et al. are right to say that dissociative symptoms are linked to unusual sleep experiences. I reason this reflects both sides of the de-differentiation equation – which exists because of the reciprocity of the neuromodulatory systems involved in the sleep–wake cycle. Although both sides of the equation would not necessarily be equivalent in magnitude, when wake is invaded by more dream-like mentation, dreaming would tend to be permeated by more wake-like, externally oriented thought. The dreams of people with schizophrenia tend to be not as numerous, shorter, less implausible, less emotional, less elaborate, and less self-oriented (Biddle Reference Biddle1963; Hadjez et al. Reference Hadjez, Stein, Gabbay, Bruckner, Meged, Barak, Elizur, Weizman and Rotenberg2003; Kramer & Roth Reference Kramer and Roth1973), with reduced emotionality both as regards dream content and in terms of the dreamer's expression of emotions (Zanasi et al. Reference Zanasi, Calisti, Di Lorenzo, Valerio and Siracusano2011). Scarone et al. (Reference Scarone, Manzone, Gambini, Kantzas, Limosani, D'Agostino and Hobson2008) reported that the dreams of people with schizophrenia are as bizarre as the dreams of normal individuals. It should be noted, however, that people with schizophrenia may report coherent dreams as bizarre because of thought disorder (Schredl Reference Schredl2011).

The invasion of more wake-like thought into REM dreaming would engender episodic memory failures because of disruption to elaborative emotional encoding. People with dissociative symptoms would also be more likely to have false memories because dream-like mentation is being elaboratively encoded during wake. For example, with dissociative symptoms, I may become convinced I had seen a house covered with sand. As regards excessive REM, this may be the brain's autopoietic response to being in a de-differentiated state. Maturana and Varela (Reference Maturana and Varela1980) emphasized the autopoiesis of living systems; such systems self-organize and self-produce. If an autopoietic system is disrupted, it “can compensate for perturbations through the realization of its autopoiesis” (Varela et al. Reference Varela, Maturana and Uribe1974). If perturbed, in this case, de-differentiated, the brain, as an autopoietic system, may organize to try to reconstitute itself (Llewellyn Reference Llewellyn2009). This organizing potential is inherent in the production of a coherence self-organized whole from constituent parts. In consequence, if REM dream encoding is disrupted through de-differentiation, then the REM period may lengthen, or intensify, to compensate. This would explain Kirov's point about why excessive REM is not linked to more successful emotional encoding.

Girard's analysis of sleep paralysis can be seen as one facet of a labile sleep–wake cycle that engenders dissociations – in this case the intrusion of REM-related motor inhibition and hallucinations into waking consciousness. In my terms, this is also a transient instance of de-differentiation between waking and dreaming – that is, a hybrid labile state. As Girard points out, in this hybrid state, hallucinations, understandably, take a different form from those in REM dreams. In sleep paralysis, there may be some conscious awareness of what are usually non-conscious hyperassociations between movement-related memories. Conscious awareness of these (during the hybrid state of sleep paralysis) engenders out-of-body experiences or felt presence.

R3.1. Redirecting the flow of emotions and reducing emotional charge over time

Emotional regulation redirects the flow of emotions from experiences (Westermann, Paulus, Müller-Pinzler, & Krach [Westermann et al.]). In the context of my arguments, this redirection is to the dream scene. This would reduce the emotional charge of the episodic memories (Desseilles & Duclos) associated in the dream scene. Emotional regulation may occur in REM because this state is characterized by the suppression of the aminergic system (Hobson et al. Reference Hobson, Pace-Schott, Stickgold, Pace-Schott, Solms, Blagrove and Harnad2003). Noradrenergic input has been linked to anxiety and stress disorders (Sullivan et al. Reference Sullivan, Coplan, Kent and Gorman1999). Therefore, revisiting emotional experiences in the absence of noradrenaline may de-potentiate them (Walker & van der Helm Reference Walker and van Der Helm2009). Similarly, van Marle et al. (Reference van Marle, Hermans, Qin, Overeem and Fernández2013) showed that administering cortisol during sleep prioritized memory processes for emotional items but reduced emotional responses to the items during retrieval in wake. Westermann et al. point out that REM sleep has been shown to increase habituation. My “Quicksand” dream could enable habituation through the identification of my previous experiences of irrational fear, particularly in the absence of noradrenaline.

Desseilles & Duclos argue that highly traumatic events are often replayed in their entirety in dreams. This may be because nothing in the individual's emotional history has prepared them for this particular trauma. In my terms, the memory fails to be associated with other previous memories and integrated into cortical networks because no previous associations are as negatively charged. The high emotional charge of the new traumatic event is not reduced through encoding. In consequence, as Westermann et al. predict, emotional load accumulates.

R3.2. REM dreaming, children's episodic networks, and infantile amnesia

Desseilles & Duclos, Kirov, and Markowitsch & Staniloiu mention REM dreaming in children. A similar logic to the foregoing paragraph may apply to infantile amnesia. My model depends on the capacity to identify non-obvious emotional associations between personal memories and portray these in mental imagery during dreaming. Young children lack this ability. First, they are still working out which associations are obvious, necessary, and destined to become part of their semantic knowledge (e.g., train – station – motion – seat – ticket) and which are non-obvious, contingent, and episodic (e.g., travelling on the mid-morning train – with my mother – we sat opposite this friendly lady – who asked me where I was going – I enjoyed telling her I was going to London). Obvious semantic associations will be encountered frequently. Hence, semantic networks are likely to form earlier than episodic. Second, young children do not have a rich history of emotional episodic experiences to which new ones can be associated. Third, children's mental imagery ability and visuo-spatial skills develop gradually. In consequence, it seems highly unlikely that very young children remember experiences through an episodic memory network with associative, integrative junctions. Children's memories may form around simple associations, but their networks are likely to be much less integrative than those of adults.

Markowitsch & Staniloiu cite evidence in support of this: The episodic memory system develops later in life; infants do not have episodic memory; children's mental imagery abilities and visuo-spatial skills increase significantly after the age of seven; during the first years of life the hippocampus (essential for making associations between episodic memories) is still immature so episodic memories cannot be formed but semantic ones can – through parahippocampal areas. There is also evidence that, in children, more adult-type dreams, featuring the agential self in dynamic, emotionally charged narratives, develop over the period from 5 to 15 years of age (Domhoff Reference Domhoff, Kryger, Roth and Dement2005; Foulkes Reference Foulkes1982; Foulkes et al. Reference Foulkes, Hollifield, Sullivan, Bradley and Terry1990). In the light of this evidence, during childhood, more sophisticated adult-type episodic networks are likely to develop gradually but, because of their different organization and structure (e.g., dynamic instantiation of junctions), may not incorporate early childhood networks. This may explain infantile amnesia.

Furthermore, if integrative adult networks do not incorporate those of early childhood, the emotional load of any traumatic early childhood experiences will not be reduced because they have not been associated with previous memories at junctions. Early traumatic childhood memories may “seek access” to the adult network in REM dreams (cf. Freud), but the associations may be too emotionally charged for hyperassociation – in REM dreams – to occur. This possibility may be what happens in the second scene of the “Quicksand” dream, in which a child/baby is thrown and disappears into the sand. These non-de-potentiated memories from early childhood are too disturbing. I awake in terror.

Kirov and Markowitsch & Staniloiu raise the question of why infants have so much REM sleep if not required for episodic memory encoding. REM sleep is needed for motor-perceptual memories (Rauchs et al. Reference Rauchs, Desgranges, Foret and Eustache2005) which, clearly, are relevant from birth. Incidentally, my argument is not that REM sleep (or even REM dreaming) only serves elaborative encoding. I agree with Kirov that REM sleep serves several functions. Also, as argued in this section, emotional encoding of episodic memories during dreaming is compatible with emotional de-potentiation (Desseilles & Duclos and Westermann et al.). Indeed, the elaborative, emotional encoding of memories and emotional de-potentiation are probably mutually constitutive – that is, they depend on each other.

I argue that episodic memory is dynamic rather than “consolidated” (in the sense of stabilized) – new junctions, formed around new nexuses of emotional associations in REM dreams, are instantiated in cortical networks during NREM sleep. Hence, the emotional load of episodic memories reduces and the emotional significance of personal memories over time, too, because memories acquire new associations as recent events are incorporated in networks.

R4.1. Non-conscious dream scenes/junctions, episodic memory, and spatial memory

To clarify, I agree with Crick and Mitchison (Reference Crick and Mitchison1983, p. 112), the function of dreaming is, “more likely to be associated with the unconscious dreaming process…rather than with the few dreams which are recalled” (p. 112). My hypothesis does not depend on dream recall into wake. However, Axmacher & Fell, along with Blagrove, Ruby, & Eichenlaub (Blagrove et al.) and Windt take issue with me on the grounds that dreams are not usually recalled into waking consciousness, whereas the AAOM occurs during wake and produces a conscious mnemotechnique. In consequence, they question whether dreaming can function mnemonically. Axmacher & Fell remark, “A mnemotechnique that sets individual memory contents into a framework that is inherently difficult to remember is paradoxical” (para. 1). To clarify, a mnemotechnique makes memories easier to retain and retrieve because they are elaboratively encoded – that is, associated. My memory framework posits that REM dream scenes encode and are used to retrieve episodic memories. To perform these dual related functions, dream scenes are not lost but retained both as junctions between episodic networks in the cortex and as hippocampal indices. On retrieval, these indices/junctions are processed by the mind/brain as indices are matched to the corresponding junctions. This processing cuts across a strict conscious/non-conscious dichotomy. Dew and Cabeza (Reference Dew and Cabeza2011) identify “porous boundaries” between conscious and non-conscious memory; the processing of non-conscious dream scenes to retrieve episodic memories into waking consciousness would be enabled by these porous boundaries.

Dream scenes do not usually reach waking consciousness. But nor are dream scenes non-conscious in the sense of “cannot be accessed.” One definition of remembered is “recalled into waking consciousness,” but another is “retained in memory.” Dreams are not usually recalled but, I argue, they are retained in memory. In their aforementioned comment, Axmacher & Fell distinguish between episodic memory contents and the organizing memory framework. Using their distinction, episodic memories are retrieved into waking consciousness. The framework (i.e., non-conscious dream scenes/junctions) is not retrieved but is processed.

One reason for the non-retrieval of junctions into wake may be because episodic memory networks evolved from those for spatial memory. Memory for spatial location is automatic, and information about location can be used as retrieval cues (Hasher & Zacks Reference Hasher and Zacks1979). During spatial navigation, information about objects placed at junctions (where a decision has to be made over where to go next) can be processed – even if that information cannot be retrieved into wake (Janzen & van Turennout Reference Janzen and van Turennout2004). Junction locations are represented (and can be retrieved) without recall into waking consciousness (Janzen et al. Reference Janzen, Wagensveld and van Turennout2007). Objects placed at junctions are automatically strongly associated with them; this differential representation (compared to objects not at junctions) would lead to faster recognition for junctions compared to other places along a navigational route (Janzen Reference Janzen2006). Faster recognition of junctions and the automatic processing of the information associated with them would greatly facilitate decisions over where to go next.. The evolutionary advantages (e.g., in fleeing predators) of this for spatial navigation are clear. If episodic memory evolved from spatial memory, the substrate for spatial memory will persist in episodic memory. In consequence, there is an evolutionary reason why episodic junctions are non-conscious. Such an arrangement would facilitate fast retrieval from multiple junctions and, thus, parallel processing in networks (see sect. R7).

Windt says I owe an explanation of why dreams are not more memorable. There are physiological causes. In terms of my arguments, if the dream scene/junction does not have to be retrieved before retrieval of the episodic micro-event memory that converges on it, retrieval is much faster. Moreover, multiple micro-events that make up a reconstructed episodic memory (e.g., see Schacter & Addis Reference Schacter and Addis2007a) can be retrieved without having to retrieve a multiplicity of mnemonic devices/dream scenes. The evidence already cited demonstrates navigation at a junction without conscious retrieval; given that episodic memory evolved from spatial memory, this evolutionary advantage may be retained in episodic networks.

Two points of clarification with regard to memory retrieval: First, Axmacher & Fell state, “The idea is that dreams trigger subsequent retrieval of parts of the episodes that are retained in them through non-conscious processes” (para. 1). This is not my position. To clarify, non-conscious dream scenes are retained both at junctions in episodic memory networks and as hippocampal indices. On retrieval, an index is matched to the corresponding junction/dream scene. The relevant episodic memory micro-event will be retrieved from one of the network paths that converge on the junction – not from the dream scene itself – which merges memory elements from all the episodic memory paths that converge on the dream scene/junction. Second, Axmacher & Fell state, “Dream memories also typically contain much less detail than does memory for experiences acquired during waking state” (para. 1). I agree, seeing this as support for my hypothesis that dream scenes are retained as indices.

An index is a sparse (or reduced) representation of material held elsewhere. The dentate gyrus is functionally specialized to instantiate sparse representations which are projected onto hippocampal region CA3 during encoding (Rolls Reference Rolls2007; see also Karlsson & Frank Reference Karlsson and Frank2008; Waydo et al. Reference Waydo, Kraskov, Quiroga, Fried and Koch2006). An index also enables access to many locations. Hence, a sparse visual index must embed multiple associations. Dream scenes are sparse as compared to visual perception of the environment in waking (Llewellyn Reference Llewellyn2011). If dream scenes are retained as hippocampal indices, this may explain their sparseness.

In “Quicksand,” I experience things I have never experienced in wake – for example, walking down a street and seeing a house completely covered in sand. The evolution of episodic memory from spatial memory may also explain why we experience the never-experienced in dreams. Gupta et al. (Reference Gupta, van der Meer, Touretzky and Redish2010) showed that, during sleep in rodents, there is replay of experienced routes but also replay of constructed, never experienced routes; the authors characterize these never experienced routes as “short-cuts.” As episodic memory evolved from spatial memory, experiencing never-experienced short-cut routes during sleep in animals may be the neural substrate from which human dreams evolved.

R4.2. REM dreams and elaborative encoding

Spoormaker, Czisch, & Holsboer (Spoormaker et al.) challenge my idea that REM dreams are elaborative encoding. They cite Spaniol et al.'s (Reference Spaniol, Davidson, Kim, Han, Moscovitch and Grady2009) fMRI meta-analysis of episodic encoding in wake which shows increased activity in hippocampus, amygdala, inferior temporal gyrus, and lateral prefrontal cortices, whereas from slow-wave sleep (SWS) to REM sleep lateral prefrontal cortices show a decrease. But my hypothesis is about emotional elaborative encoding, a process that identifies non-obvious emotional associations between episodic events. In Spaniol et al.'s meta-analysis of encoding/retrieval during wake, in 20 of the 26 studies the stimulus was a word – none of the 26 studies tested for encoding episodic events. In terms of regional brain activation, emotional associative encoding of episodic events is likely to differ from encoding words during wake. Indeed, in an fMRI study, McDermott et al. (Reference McDermott, Szpunar and Christ2009) showed that laboratory word-based studies to test episodic memory shared very few regions of overlap with studies where participants were asked to remember real-life events. Hence, it is unsurprising that fMRI studies of encoding words during wake do not show all of the same regional activations as studies of REM sleep.

Spoormaker et al. also cite the Maguire et al. (Reference Maguire, Valentine, Wilding and Kapur2003) study of superior memorizers who demonstrated preferential engagement of the retrosplenial cortex (RSC) during encoding. In a review, Vann et al. (Reference Vann, Aggleton and Maguire2009) found that the RSC is involved in episodic memory and imagination, specifically in translating allocentric viewpoints to egocentric. In the study by Maguire et al., the superior memorizers translated digits into images; this could account for the engagement of the RSC. Spoormaker et al. equate the RSC and the posterior cingulate cortex, arguing that the posterior cingulate cortex shows a decrease from SWS to REM sleep. But this equation is not justified. The Vann et al. review showed the importance of distinguishing the function of the RSC from other parts of the posterior cingulate region, arguing that “the RSC is most likely to be involved in hippocampus-dependent functions” (p. 793).

R4.3. Dream scenes and the method of loci

Nielsen presents a detailed case against the method of loci (MoL) operating in dreams – in exactly the same way as it does in contemporary mnemonic techniques (see also Dresler & Konrad). To clarify, my hypothesis does not rely on dream scenes following each other along a personally familiar route (or rote memorized path as in a memory palace); this would pre-determine the order in which episodic memories are retrieved and, thus, greatly reduce memory flexibility. Nielsen notes that the MoL is a favoured technique amongst memory champions. This is because contemporary mnemonic competitions often test the ability to remember serially ordered material and recall that material in the same order – as noted by Dresler & Konrad.

The original uses of ordering in the AAOM were not to recover lists of items but to increase memory flexibility so that oratory digressions were possible. The AAOM focussed mainly on the encoding of material needed for what were, often, extempore events, such as sermons, lectures, or disputations before live audiences (Carruthers & Ziolkowski Reference Carruthers, Ziolkowski, Carruthers and Ziolkowski2002, p. 3). On such occasions, it was important, clearly, to be able to extemporize. Using the AAOM, in classical antiquity, speakers could digress, to enlarge on points, because they were always sure of “where they were” in their speeches, “not in the manner of a parrot (which reciting mindlessly, never knows ‘where it is’) but in the manner of an experienced harbour pilot recalling landmarks” (p. 5).

With respect to retrieval from places, my argument is analogous to the original use of the AAOM, resting on episodic micro-event memories being recovered from places (i.e., junctions/retained dream scenes in the cortex) to reconstruct memories in flexible ways. In consequence, the question is how the ordering of these places (cortical junctions/hippocampal indices) is structured to enable flexibility. As Nielsen remarks, this orderingof dream scenes may be emotionally associational rather than spatially associational – which would be anticipated if episodic memory evolved from spatial memory. I argued, “The associations forged in REM dreaming may originally have encoded spatial landmarks but through evolutionary time became episodic” (sect. 4.2.5, para. 4). In the “Quicksand” illustration, my remote “fear of quicksand” memory has a subordinate role in scene 2 (the cloth–shroud–death–sand association to the house) but assumes a central role in scene 3, where a child/baby is thrown across the sand and disappears. Thus, it may be possible to trace emotionally salient associations across dream scenes through assuming that a peripheral (to the particular scene) emotional association in one scene becomes dominant in the next.

Dresler & Konrad suggest that mnemonic techniques linked to ordering (spatially or emotionally associational) are less relevant for episodic memories because such memories are already in sequential order. But the key attribute of human episodic and semantic memory is flexibility – so that memories can be used in future situations that differ from the original past context (e.g., see Reber et al. Reference Reber, Knowlton and Squire1996; Schacter & Addis Reference Schacter and Addis2007a; Schacter et al. Reference Schacter, Addis and Buckner2007). Linked to this flexibility is the ability to identify non-obvious patterns among seemingly different events.

In the AAOM, to enhance memory flexibility, the to-be-remembered material is split up into divisiones or distinctiones (Carruthers and Ziolkowski Reference Carruthers, Ziolkowski, Carruthers and Ziolkowski2002, p. 4). Similarly, I argue that in REM dreams episodic memories are split into micro-events through pattern separation in the hippocampus. This may explain the Grosmark et al. (Reference Grosmark, Mizuseki, Pastalkova, Diba and Buzsáki2012) finding of synaptic downscaling in the hippocampus during REM sleep (see Kirov). Pattern separation facilitates flexibility because micro-events can be assembled differently. For example, in “Quicksand,” the micro-event memory of photographing bridges was an aspect of the U.S. holiday memory, but, equally, it could be retrieved in the context of memories of challenging photography experiences or feature in a discussion about bridge construction. In the latter two cases, the U.S. holiday context is irrelevant. In any case, who wants to listen to me reminiscing about the whole of my U.S. holiday? Freeing episodic micro-events from their whole memory context – to enable flexibility – greatly enhances individual learning and social interaction. Given this flexibility, how does ordering of the hippocampal index work in retrieval? I address this next.

R5.1. The ordering of the hippocampal index

On retrieval, hippocampal indices are matched against cortical junctions, and the ordering of the hippocampal index enables this matching. Thus, cognitive maps in the hippocampus (O'Keefe & Nadel Reference O'Keefe and Nadel1978) represent junctions/landmarks in the cortex. Cheng & Werning agree that hippocampal place cells may represent spatial locations in the cortex. I argue that hippocampal indices would be activated by place cell firing. However, the memory output from any input (external stimulus or internal thought) may not be pre-determined solely by the ordering of REM dream scenes. The reason may relate to the function of NREM dreams.

R5.2. What is the function of NREM dreams?

Cicogna & Occihionero reason that associative areas are similarly activated in REM and NREM and that the function of both REM and NREM dreams may be the forging of associations.

I argue that emotional hyperassociation is the basis for REM dreams and also that emotional association (not hyperassociation) drives the ordering of dream scenes and, thus, the ordering of hippocampal indices in the retrieval of emotional episodic memories. The function of NREM dreams may be related to the associative ordering of the hippocampal index.

I noted a contrast between REM and NREM sleep with regard to association. Stickgold et al. (Reference Stickgold, Scott, Rittenhouse and Hobson1999) showed that, “contrary to normal pattern of priming, subjects awakened from REM sleep showed greater priming by weak primes than by strong primes…. In contrast, strong priming exceeded weak priming in NREM sleep” (p. 182). They argue that weak priming immediately after awakening from REM sleep may reflect the hyperassociative nature of REM dreams. NREM dream associations may have greater similarity to the more logical associations made in wake. These NREM associations may be analogous to “cross-references” in the hippocampal “index.”

I illustrate this with a dream that may be from NREM. (Details and hypothesized memory sources are in the Appendix.) Briefly, the dream occurred during an afternoon nap after arriving at a motel. My recent memories/concerns are these: I plan to swim later and have a drink in the motel bar where I saw a man sitting with his back to me. For a woman travelling alone, this bar looks OK. I am concerned that I have not brought a sufficient supply of contact lenses for my trip; and I am attracted to K.C., who, I think, is split from his partner, like me. My remote memories/concerns are these: an emotional problem with my youngest son and a weekend with him in Barcelona; and a conference in Barcelona where, on arrival, I was confused about location.

Scene 3 is the most detailed. I suggest this dream scene is from the previous REM period (see the Appendix). In NREM, this scene is being instantiated as a junction, and the scene is again consciously experienced, albeit in a different neuromodulatory milieu. Consequently, like a dream report in wake, a NREM dream experience of a REM dream will differ.

As instantiation occurs, the hippocampus identifies associations between scene 3 and other more remote memories/concerns at other cortical junctions: My attraction to K.C., who goes frequently to Barcelona (scene 5); and arriving at a conference venue in Barcelona and walking back to my hotel from this conference (scenes 4 and 6). Scenes 1 and 2, being with my sisters by the sea and at the pool, when, because I was not wearing contact lenses, I could never see well, and scene 2, the kiss, may be cross-references to the previously instantiated junction, which will be related to Barcelona. All of these NREM associations may be analogous to cross-references in the hippocampal index; they are associations between junctions rather than associations within junctions, as in REM dreams:

Barcelona, Bar-sea-lona, Bar-see-loner: a dream illustration of more thought-like associations in simple, shorter scenes (1, 2, and 4–6) as would be engendered by strong primes, in contrast to the more complex, less obvious associations in scene 3, as would be reflected in weak primes.

• Scene 1 (Swimming with my sisters was always associated with not seeing well, so “swimming with sisters” would be a strong prime for “not seeing well”). I'm lying in the sunshine. My sisters are with me. It's a beautiful day, the light is perfect, but everything is a bit blurry. Why can't I see very well? I am by a curvy swimming pool, but, at the same time, it feels as though I'm by the sea. (Cross-reference from the previously instantiated junction.)

• Scene 2 (I am attracted to K.C., so “K.C.” would be a strong prime for “kiss”). K.C. is kissing me, a long, deep kiss (this may be in scene 1 or there may not be a location for this kiss). (Cross-reference from the previously instantiated junction.)

• Scene 3 (Complex associations between emotional issues with my youngest son, a weekend with him in Barcelona, finding more contact lenses, and spatial directions which would be engendered by weak primes). I am walking along a street in the sunshine with an expanse of long, flat water on one side. In the distance, I see D. (my youngest son) and feel worried. He stands with, but slightly apart from, some girls. I call to him to hurry up and join me, but he doesn't seem to want to; at any rate, he stays right where he is. I feel lonely and wonder why he doesn't come to me. I probably have to accept that he would rather be with girls now. I pick up a huge, shiny, elongated triangle off the ground. It was pointing up to where D. is. What is this? It is most peculiar. Whilst being a triangle, it is also transparent, reflective, floppy, and curved. It is concave when looked at from beneath and convex when viewed from the top. I am unsure of what to do with this object. (Instantiation of hippocampal “Barcelona” index/junction.)

• Scene 4 (Arriving at a conference in Barcelona, I thought I was in the wrong place, so “arriving at a conference in Barcelona” would be a strong prime for “wrong place”). I have reached a crossroads and I'm not sure of the way. I think that it must be downhill. Suddenly the crossroads dissolves into a large, paved, shiny white open space/place. Where am I? (“Barcelona” cross-reference.)

• Scene 5 (I am attracted to K.C., and he is substituted for the man in the bar so “man in a red jacket in Barcelona” would be a strong prime for “K.C”). I enter a bar. I see K.C. from behind, sitting at the bar. He is wearing his red jacket. He is chatting with friends (men and women) and doesn't notice me. I would like to join him – but feel that I cannot. (“Barcelona” cross-reference.)

• Scene 6 (Confident of directions, I return to my Barcelona hotel after the conference, so “returning to Barcelona hotel” would be a strong prime for “confident of directions”). I am back again on the street-alone, walking downwards but feeling quite serene. (“Barcelona” cross-reference.)

This hypothesis of the associative function of NREM dreams implies increased functional cortico-cortical connectivity in stage 2 NREM as junctions are instantiated and associated with already instantiated junctions. The instantiation of new junctions would be facilitated by prior network reorganization during SWS. Spoormaker et al. (Reference Spoormaker, Schröter, Gleiser, Andrade, Dresler, Wehrle, Sämann and Czisch2010) demonstrated the development of a large-scale functional brain network during human NREM sleep and confirmed the Massimini et al. (Reference Massimini, Ferrarelli, Huber, Esser, Singh and Tononi2005) finding of a breakdown in cortical connectivity in SWS.

Spoormaker et al. cite Baylor and Cavallero (Reference Baylor and Cavallero2001) who state that more episodic memory traces are recovered from NREM dreams than from REM. I agree this would be the case. In “Barcelona,” my memories/concerns are relatively transparent. I enter a bar (scene 5), am lying by a pool, not able to see (scene 1), and pick up a contact lens (scene 3). Episodic elements are more difficult to identify in REM dreams like “Quicksand” because the memories are hyperassociated rather than only associated, as is the case in “Barcelona.” Hyperassociation renders memory elements less identifiable.

Hassabis et al. (Reference Hassabis, Chu, Rees, Weiskopf, Molyneux and Maguire2009) argue that the hippocampal neural code (index) has functional organization. If hippocampal indices are related thematically (both emotionally between REM dream scenes and more linear-logically between NREM dream scenes), multiple indices can be called up in succession. This sequential ordering relies on identifying associations between patterns through recurrent connections in CA3 (for a review, see Bird & Burgess Reference Bird and Burgess2008). Thus, whether in direct response to internal hippocampal indexical representations or through external stimuli which are matched against these indices, several cortical junctions can be identified that enable access to the several components of a whole episodic memory (the U.S. holiday) or, through cross-references, a series of associated micro-events (e.g., several of my photography experiences or several personal associations with Barcelona). This is reconstruction (cf. Bartlett Reference Bartlett1932; also see Neisser Reference Neisser, Gladwin and Sturtevant1962; Schachtel Reference Schachtel1947). Entire episodes cannot be immediately “read off.” Thus, the hippocampal index (with both entries [REM dream scenes] and cross-references [NREM dream scenes]) ensures both ordering and flexibility when a stimulus triggers memory retrieval.

For example, suppose I meet someone at a party (where social chit-chat is required) who also knows Barcelona well. I have several episodic associations with the place (memory flexibility) because the “Barcelona” hippocampal index has cross-references. I can match one of these to the interests of the party companion – to start a conversation. If the person is an academic, I can talk about the conference. If the person happened to go there with one of their children, I can chat about my weekend with my son. The stimulus “Barcelona” does not trigger some predetermined extended episodic memory (in the way the MoL can be used to generate a list). This would be a social faux pas – because I would have failed to relate my interests to those of my new acquaintance. Equally, the stimulus “Barcelona” could enable learning because I could associate the contact-lens crisis with a series of other travelling-with-contact-lens disasters to work out how to avoid them.

Cicogna & Occihionero recommend that future research address the question of the recall, during dreaming, of previously encoded dream episodes/scenes because dreams, retrieved into dreaming, may serve as a basis for generating the dreams that follow them. In the foregoing, I suggest this is exactly what happens in the NREM dreams that follow REM dreams. The instantiation of a REM dream as a junction in NREM again triggers the REM dream, albeit experienced somewhat differently. The hippocampus then identifies associations between the REM dream scene/junction in NREM and the dream scene/junctions already instantiated in the cortex. This associative function in NREM maintains ordering in the hippocampal index, but through incorporating cross-references also ensures the flexibility of episodic memory.

Hence, I agree with Dominey that my hippocampal indexing model of memory retrieval is based not on the reconstruction of a single state trajectory but, flexibly, on the reconstruction of one (or more) of several state trajectories. This is ensured by cross-references in the hippocampal index. Dominey describes hippocampal indices as “snapshots.” Again I agree with this because in modelling the cortex as a dynamic network memory system my model concurs with his.

R5.3. What is the function of the different sleep stages across the night?

Axmacher & Fell, Blagrove et al., and Deliens, Schwartz, & Peigneux (Deliens et al.) point out that if REM dreams are instantiated as junctions in NREM, then, logically, NREM sleep should follow REM (and some does), but most NREM sleep occurs in the first half of the night.

This objection disregards my statements about the earlier instantiation of semantic material in NREM:

If semantic material is instantiated in networks in earlier periods of NREM sleep, it may be necessary to distinguish between semantic and episodic memory when considering memory processes across sleep stages. This is in line with Cicogna & Occihionero's arguments about the intricacy of sleep stages and the need to distinguish them carefully. In a review, Fogel and Smith (Reference Fogel and Smith2011) suggest that complex memory tasks may require REM sleep, whereas simpler ones and/or tasks already fully encoded during wake may only require NREM sleep. In my terms, this implies that semantic knowledge that is fully encoded in wake is instantiated during early-night NREM, whereas episodic memories that are not fully encoded during wake undergo further emotional encoding in REM. Therefore, semantic material (or knowledge) encoded in wake may be instantiated and integrated into semantic memory networks during the first half of the night whilst episodic memories may be instantiated and integrated into episodic networks in the second half of the night. This would explain why early-night NREM dreams are less emotionally charged and more thought-like, whereas later-night NREM dreams (that follow REM) become more emotional and more REM dream-like as the night progresses.

I noted that Clemens et al. (Reference Clemens, Fabó and Halász2005) found a correlation between time in NREM sleep and visuo-spatial memory, as demonstrated by the retention of faces. This may be consistent with the earlier instantiation of verbal memory junctions through slow spindles and the later instantiation of visuo-spatial memory junctions through fast spindles after the onset of REM dreaming. The topographic distribution of spindles may also be relevant because slow (verbal) spindles have a more frontal, anterior distribution, whereas fast (visuo-spatial) spindles have a centro-parietal, more posterior distribution (for a review, see Fogel & Smith Reference Fogel and Smith2011). This implies that the shift in dream content overnight (more thought-like to more emotional) reflects movement across memory networks from anterior to posterior. If this is correct, when I recorded “Barcelona” – after waking from a 4-hour sleep during the day – I was approximately halfway through this cycle. I may have captured a NREM dream after the first short REM period. This would explain the more semantic nature of this dream; up until this point NREM dreams had been associating semantic material. Specific REM periods may be linked to particular tasks (Smith Reference Smith, Moffitt, Kramer and Hoffman1993; Smith et al. Reference Smith, Nixon and Nader2004). For example, Stickgold et al. (Reference Stickgold, Whidbee, Schirmer, Patel and Hobson2000) found that, on a visual discrimination task, performance was correlated with the amount of REM sleep in the last quarter of the night. This would make sense if REM dreams become more visuo-spatially complex over the course of the night. The first short REM period may be encoding memories/concerns that are related to semantic material, like the word “Barcelona.” A significant aspect of human memory is the recall of associations to people, places or, even, symbolically significant objects that occurred at different times, and the first REM/NREM cycle may support these types of memories.

Drawing on Braun et al. (Reference Braun, Balkin, Wesensten, Carson, Varga, Baldwin, Selbie, Belenky and Herscovitch1997) and Maquet et al. (Reference Maquet, Peters, Aerts, Delfiore, Degueldre, Luxen and Franck1996), Spoormaker et al. state that the REM sleep recruits only the anterior part of episodic memory networks, but Braun et al. investigated the first REM period only, during which, if the aforementioned temporal shift from anterior, more semantically associated, to posterior visuo-spatial, more emotionally intense, hyperassociated REM dream content is correct, only the anterior aspect would be recruited. However, later REM periods would recruit posterior parts of the episodic network. Maquet et al. looked at two REM periods, but their analysis averaged out cortical activations specific to these periods. Therefore, Spoormaker et al.'s assumption about the recruitment of only the anterior part of episodic networks during REM sleep is somewhat risky.

R6.1. Fit with current network models

Mattei argues that my ideas of how memories are encoded in networks are supported by the current “hodotopic model” of brain function (para. 3), which relies on local recurrent activity and parallel processing driven by non-linear (chaotic) dynamics. Cross-references in the hippocampal index are non-linear and may further support parallel processing.

When connectionist computational models, to explore parallel, distributed processing, are built, they include hidden units which mediate between the inputs and the outputs. Older connectionist models lacked these hidden units. In consequence, they modelled only stimulus-response type behaviour. With hidden units, the model becomes non-linear and can solve more complex problems and tackle an increased range of problem types (Rumelhart & McClelland Reference Rumelhart and McClelland1986). The hidden units embed deep associations between the inputs and outputs through abstracting away from their more superficial features. With hidden units, the model can also better differentiate among the inputs to enable more nuanced outputs.

I suggest these hidden units in connectionist models are analogous to the non-conscious junctions/dream scenes that are processed during retrieval of episodic memories. The title of Mattei's commentary refers to the secret at the crossroads. The associations embedded in junctions/crossroads between episodic micro-events abstract away from the more superficial aspects of the micro-events. For example, the central non-obvious (secret) association hidden in the various micro-events that converge on the “Quicksand” junction is irrational fear. The “Quicksand” junction also enables differentiation between the micro-events because this irrational fear association is non-obvious – that is, the phone call to my eldest son, fear of quicksand, photographing bridges in the U.S., and my middle son's hearing problem are, superficially, very different.

R6.2. The nature of associative network junctions/dream scenes

Cheng & Werning point to further physiological evidence for aspects of my hypothesis, but they assume an argument against me by contending REM sleep generates “invariant object representations.” This puzzled me because a central association between episodic micro-events is invariant. For example, after abstraction from more superficial features, irrational fear is an invariant feature of the micro-events in “Quicksand.” I was also surprised that Deliens et al. characterize REM dream associations as de novo. Rather, irrational fear in “Quicksand” is identified through pattern identification, which is novel (rather than de novo) because it is non-obvious. In wake, I would be unlikely to see a pattern between the phone call to my eldest son, fear of quicksand, photographing bridges in the U.S., and my middle son's hearing problem. To the waking mind, these seem disparate. Perogamvros and Schwartz (Reference Perogamvros and Schwartz2012) argue that dopaminergic activation during REM sleep favours unusual (non-obvious) associations. The irrational fear association is unusual, given the nature of the micro-events involved.

Pearlmutter & Houghton argue that memory processes during sleep are secondary – the primary purpose of sleep is tuning for criticality. But, surely, sleep, like wake, has a multitude of purposes. I agree with their tuning-for-criticality idea when they say:

Sleep provides an opportunity to tune cortical networks to prevent runaway oscillations. I argue junctions achieve this because they are sites of inhibition that avoid runaway oscillations. Pearlmutter & Houghton say that super-critical, uncontrolled behaviour would result if the network was not fully tuned during sleep. Creativity is super-critical behaviour (see sect. R2): If tuning is impeded because of increasing de-differentiation between wake and sleep, uncontrolled behaviour (e.g., psychosis or, ultimately, schizophrenia) may result. I noted spindle impairment in early psychosis and reduced spindle activity in schizophrenia.

R7.1. Tests suggested by the commentators

Dominey remarks that indexation is potentially testable through re-injecting indices into the neocortex to enable the unfolding of an episodic sequence. Cicogna & Occihionero stress the significance of dreaming in different sleep stages. I propose: NREM stage 2 dreams in the first half of the night enable the formation of junctions in semantic networks; the first REM period hyperassociates semantic and episodic memories, the next NREM period enables the formation of junctions in anterior episodic networks and associations between new junctions and ones already instantiated; and later REM periods hyperassociate emotional episodic memories that, in NREM, enable the formation of junctions in posterior episodic networks and associations between junctions. This is testable through careful analysis of dream associations along with fMRI of regional brain activation. Dresler & Konrad undertook a simple qualitative test of my hypothesis through asking memory champions whether mnemonics feels like dreaming. The answer was equivocal. A single question may be inconclusive, but several may work – for example, enquiring as to the similarity between the mnemonic AAOM product and a dream. Dresler & Konrad demonstrated that, after memory champions use mnemonic techniques, their REM sleep duration and REM density do not differ from those of mnemonically naive controls. However, methodological considerations may be relevant. Dresler & Konrad state that their test required the encoding of declarative information, but this could have been episodic memories or semantic material. If the latter, changes in REM sleep parameters would not be expected – rather the impact would be on NREM sleep. Moreover, it is uncertain what impact successful elaborative emotional encoding of recent episodic memories during wake would have on subsequent REM sleep. Would there be less to encode in REM dreams? However, testing my hypothesis with memory champions is important. Another testing avenue, which is implicit in the Markowitsch & Staniloiu commentary, is testing whether more creative individuals, like writers, are encoding their episodic memories in wake and, therefore, would show an enduring neurobiological shift to a state of mind/brain more REM dream-like.

R7.2. Possible evidence against the hypothesis?

Solms argues that there is evidence against my hypothesis on two grounds: First, the loss of dreaming in patients with cerebral lesions, investigated by Solms (Reference Solms1997) and Yu (Reference Yu2006), showed no demonstrable effect on episodic memory. But, so far as I can ascertain, neither Solms nor Yu tested for episodic memory loss, so Solms's conclusion seems somewhat cavalier. On his memory testing, Solms (Reference Solms1997) remarks:

Over and above these tests, Solms (Reference Solms1997) remarks, “I assessed remote memory informally by questioning patients about significant public events and popular television programmes” (p. 258). Such questioning does not test patients' memory for their remote episodic experiences; it probes their remote semantic memory. Moreover, Yu (Reference Yu2006) comes to a rather different conclusion from Solms – he states that “it is perhaps unsurprising to find that most patients who have ceased dreaming also experience memory deficits” (p. 192). Yu describes his tests as follows: “The neuropsychological battery used in this study is composed of three primary parts, which measure executive functions, visual memory and verbal memory respectively. These three components constitute a critical test for the potential association between dream cessation and memory dysfunction” (p. 192). Unfortunately, these critical tests did not seem to include appropriate testing for patients' episodic memories. In this context, as mentioned previously, Solms and Turnbull (Reference Solms and Turnbull2002) observe that non-dreaming patients become “aspontaneous, inert and apathetic” (p. 312). This would be anticipated if there is episodic memory loss, because this would erode the patients' ability to imagine, plan, and enact the future (see sect. R2).

Second, Solms comments that bilateral hippocampal lesions, causing total loss of episodic memory, have no demonstrable effect on the occurrence of REM-like dreams. But why does he think this relevant? My hypothesis is not concerned with mechanisms that control the occurrence of the dream state but with dreaming functionality. Mere occurrence does not ensure functionality. For example, the wake state occurs in people with schizophrenia, but their waking functionality is severely impaired. Similarly, Solms argues that dreaming is not controlled by hippocampal mechanisms. If, by “controlled,” Solms means generated in the sleep cycle, I agree; but if he implies that hippocampal mechanisms are not involved in dream physiology and phenomenology and, hence, possibly in dream function, I disagree. At any rate, if Solms's lesion work is continuing and if appropriate episodic memory tests are applied, this work is significant with regard to testing my hypothesis (but see Schredl, who identifies some methodological difficulties with lesion work).

Markowitsch & Staniloiu point out that testing for relationships between episodic memory and dreams will depend on how episodic memory is conceptualized and the nature of the testing paradigms employed. This is exemplified in the previously mentioned McDermott et al. (Reference McDermott, Szpunar and Christ2009) study which demonstrated that laboratory word-based studies to test episodic memory shared very few regions of overlap with studies where participants were asked to remember and re-experience real-life events.

Axmacher & Fell say there is little evidence that dreaming is significant for memory, but how could there be significant evidence in view of the scant research to date? As Nielsen points out, though, there is actually a recent upsurge of interest in the role of dreaming in memory processes (e.g., see Smith Reference Smith2010; Wamsley & Stickgold Reference Wamsley and Stickgold2011; Wamsley et al. Reference Wamsley, Tucker, Payne, Benavides and Stickgold2010). This is unsurprising given the escalating experimental evidence on the links between sleep and memory. I would welcome the opportunity to be involved in experimental approaches to test my ideas and those of others so that, as Hobson puts it, I “stand up” as an experimentalist. The experimental approach is fundamental to science. On the other hand, when I entered the field of dream research, I was surprised at the dearth of theory. I believe passionately in the complementary contributions of theory and data so that data can support or challenge theory and theory can situate and give meaning to data. To paraphrase Immanuel Kant, theory without data is empty, but data without theory is blind.