Anaphoric distance dependencies in visual narrative structure and processing

Linguistic syntax has often been claimed as uniquely complex due to features like anaphoric relations and distance dependencies. However, visual narratives of sequential images, like those in comics, have been argued to use sequencing mechanisms analogous to those in language. These narrative structures include “ refiner ” panels that “ zoom in ” on the contents of another panel. Similar to anaphora in language, refiners indexically connect inexplicit referential information in one unit (refiner, pronoun) to a more informative “ antecedent ” elsewhere in the discourse. Also like in language, refiners can follow their antecedents (anaphoric) or precede them (cataphoric), along with having either proximal or distant connections. We here explore the constraints on visual narrative refiners created by modulating these features of order and distance. Experiment 1 examined participants ’ preferences for where refiners are placed in a sequence using a force-choice test, which revealed that refiners are preferred to follow their antecedents and have proximal distances from them. Experiment 2 then showed that distance dependencies lead to slower self-paced viewing times. Finally, measurements of event-related brain potentials (ERPs) in Experiment 3 revealed that these patterns evoke similar brain responses as referential dependencies in language (i.e., N400, LAN, Nref). Across all three studies, the constraints and (neuro)cognitive responses to refiners parallel those shown to anaphora in language, suggesting domain-general constraints on the sequencing of referential dependencies.

Visual language Syntax Narrative Anaphora Distance dependencies ERPs A B S T R A C T Linguistic syntax has often been claimed as uniquely complex due to features like anaphoric relations and distance dependencies.However, visual narratives of sequential images, like those in comics, have been argued to use sequencing mechanisms analogous to those in language.These narrative structures include "refiner" panels that "zoom in" on the contents of another panel.Similar to anaphora in language, refiners indexically connect inexplicit referential information in one unit (refiner, pronoun) to a more informative "antecedent" elsewhere in the discourse.Also like in language, refiners can follow their antecedents (anaphoric) or precede them (cataphoric), along with having either proximal or distant connections.We here explore the constraints on visual narrative refiners created by modulating these features of order and distance.Experiment 1 examined participants' preferences for where refiners are placed in a sequence using a forcechoice test, which revealed that refiners are preferred to follow their antecedents and have proximal distances from them.Experiment 2 then showed that distance dependencies lead to slower self-paced viewing times.Finally, measurements of event-related brain potentials (ERPs) in Experiment 3 revealed that these patterns evoke similar brain responses as referential dependencies in language (i.e., N400, LAN, Nref).Across all three studies, the constraints and (neuro)cognitive responses to refiners parallel those shown to anaphora in language, suggesting domain-general constraints on the sequencing of referential dependencies.
This sequence shows Lucy reaching into a bag (panel 1) sitting next to Charlie.She then tosses a candy (panel 2) as Charlie watches (panel 3) and we see the candy in the air (panel 4).Snoopy then jumps between them to grab the candy (panel 5) leaving them stunned by his interruption (panel 6).While understanding these events alone can be complex, consider specifically the relationship between panels 2 and 4. Panel 4 shows a zoom of the same information in panel 2-a refiner-a part-whole indexical relationship (Bateman & Wildfeuer, 2014a;Cohn, 2015), which also maintains co-reference in that the contents of both panels refer to the same entity (Klomberg, Hacımusaoglu, Lichtenberg, Schilperoord, & Cohn, 2023).Indexical relationships maintaining coreference are also how anaphora in language has long been characterized (Bühler, 1934)-i.e., referential dependencies like her, it, or themselves refer to other units of the sentence or discourse (Chomsky, 1981;Culicover & Jackendoff, 2005;van Dijk & Kintsch, 1983).
In addition, these co-referential panels are separated at a distance, indicating that that the proximal relations between panels are not enough to account for their relationship.Distance dependencies-the connections between non-adjacent units (here: panels)-have also long been acknowledged as a fundamental trait of syntactic structure (Chomsky, 1965;Gibson, 2000).In fact, the sequence in Fig. 1 demonstrates a consistent pattern that occurs across comics, and has been shown in corpus analyses to vary cross-culturally (Cohn, 2019), suggesting diverse standardization.
Here, we explore this particular refiner pattern and its family of related constructions by drawing analogies with two paradigmatic features of linguistic structure: distance dependencies and anaphora.We will first describe the constructs involved with these sequences, then show across three experiments that constraints operating on visual narrative sequencing share architectural and neurocognitive analogous to those in structures of language.

Basic narrative structure
To comprehend a visual sequence, a reader must extract the relevant cues from each panel and then access the corresponding semantics.For Fig. 1, this involves recognizing that there are characters (Charlie Brown, Lucy, Snoopy) and that they undertake events (tossing candy, jumping, biting).All of this information is incorporated into a growing mental model for understanding this visual discourse (Cohn, 2020b;Loschky, Magliano, Larson, & Smith, 2020), consistent with the processes involved in understanding textual discourse (Cohn & Magliano, 2020;Gernsbacher, 1990;McNamara & Magliano, 2009;Zwaan & Radvansky, 1998).With each successive image, a reader must update the mental model for information that changes or shifts across panels (Cohn, 2020b;Loschky et al., 2020), such as shifts between characters (panels 2 and 3) or events (panels 1, 2, 5, 6).
While much work on visual narrative comprehension has looked at general properties of meaning-making across image sequences (e.g., Bateman & Wildfeuer, 2014a, 2014b;Gernsbacher, 1985;Magliano, Larson, Higgs, & Loschky, 2015;Magliano, Miller, & Zwaan, 2001;McCloud, 1993), sequential images use specific constructional patterns (Bateman & Schmidt, 2012;Branigan, 1992;Cohn, 2015;Metz, 1974).To account for these patterns, Visual Narrative Grammar (VNG) posits a "narrative grammar" that persists in parallel with semantic structures using similar architectural principles as a syntactic structure, though operating at a discourse level of information (Cohn, 2013(Cohn, , 2015)).This separation between narrative and semantic levels of representation parallels such a division in language (Jackendoff, 2002), and is supported by (neuro)cognitive experimentation (Cohn, 2020b).While VNG shares the "grammatical" orientation to narrative as prior "story grammars" which were inspired by Chomskyan generative grammar (Mandler & Johnson, 1977;Rumelhart, 1975), those approaches were limited in both their theoretical formalisms and their experimental methods.In contrast, VNG is consistent with linguistic models of construction grammar which posit grammatical patterns are stored in memory as declarative lexical schemas, rather than as procedural rules (Culicover & Jackendoff, 2005;Goldberg, 1995), and VNG has been supported by contemporary psycholinguistic methods (for review and discussion, see Cohn, 2020b).
A canonical narrative sequence progresses through several basic states (Table 1a).An Establisher may start a sequence, often by introducing the characters, as in the first panel of Fig. 1.An Initial then begins their interactions, such as depicting a preparatory action (Lucy tossing the candy), which then climaxes in the Peak (Snoopy unexpectedly grabbing the candy).A Release then diffuses the tension of this interaction, often as the coda or aftermath of an event (Lucy and Charlie looking bewildered).These categories comprise Fig. 1.A visual sequence using a refiner displacement pattern along with its diagrammed narrative structure.Peanuts is © Peanuts Worldwide LLC.
N. Cohn et al. a canonical narrative schema, which is stored in long-term memory as a sequence pattern (for review, see Cohn, 2020b).Because this schema is inherently recursive, narrative categories can apply both to individual panels and to constituents of panels, allowing for center-embedding, long distance dependencies, and other constructional patterns (Cohn, 2013(Cohn, , 2015)).Segments are combined using "unification"-a process of assembling schematic pieces stored in memory into a larger structure (Hagoort, 2017;Jackendoff, 2002).
The canonical narrative schema can also be modified.In Fig. 1, panels 2 and 3 each draw attention to a single character (Lucy, Charlie), rather than showing them together in the whole scene.Because these two panels can combine in a single Initial panel, and could be rearranged in their positions (Cohn, 2015), they both function as Initials, united by a conjunction schema that combines similar narrative categories within a constituent of that same category (Table 1b).This is similar to conjunction in sentences where grammatical categories conjoin within a phrase of the same type, such as two nouns within a noun phrase (e.g., bread and butter).We will refer to characters appearing in these panels by letters-character "A" and character "B"-making this conjunction structured as [A -B].This particular conjunction also creates the need for a spatial inference, since the characters are not viewed in a shared location.Indeed, a diagnostic for this construction is to substitute panels involved in conjunction for a single panel that contains all those characters (Cohn, 2015).This construction has been called environmental-conjunction (e-conjunction) because it is a narrative conjunction which motivates an inference of the semantic environment.The spatial inference in semantics is co-indexed to narrative structure, notated by a subscript "e" to the constituent-level narrative category (Fig. 1).
Other modifiers repeat information.In Fig. 1, panel 4 zooms in on information from a preceding panel (the arm tossing candy).While zoomed-in framing can be used at any narrative category, refiners like these specifically provide a focal view of information in a "head" panel (e.g., Lucy's hand and the candy) (Cohn, 2015).We will use upper-and lowercase letters to depict the relationship between a character ("A") and its refiner ("a"), making a head-modifier relationship [Aa].A diagnostic for identifying refiners is whether an inset panel (Foulsham & Cohn, 2021)-a panel within a panel-can be drawn around the "refined" information within the head and be similar to the contents of the refiner.
Refiners are "anaphoric" to their heads in that they co-refer, just as less informative units (pronouns) referentially index their more informative antecedents (full nouns) (Bühler, 1934).Likewise, a refiner and head exhibit an indexical part-whole relationship (Bateman & Wildfeuer, 2014a;Cohn, 2015).Structural features characteristic of anaphora in language may also persist for refiners in visual narratives, consistent with findings that they use similar processing mechanisms (Cohn, 2020b;Gernsbacher & Robertson, 2004;Loschky et al., 2020), including between anaphoric relationships and refiners (Coopmans & Cohn, 2022).For example, like anaphora in language, "anaphoric refiners" follow their antecedents, while "cataphoric refiners" precede their antecedents, as reflected in the head-modifier schema for refiners in Table 1c.We return to this analogy below.
Instead of grammatical patterns, another approach drawing from linguistics has focused on the meaningful relations between panels (Bateman & Wildfeuer, 2014a, 2014b), implementing theories of discourse semantics (Asher & Lascarides, 2003).Under this analysis, panels or constituents with different characters (e.g., e-conjunctions) display a coherence relation of contrast, shifting between different elements of the scene, while zooming panels (refiners) use a relation of elaboration or detail, expanding on information in a different unit (Bateman & Wildfeuer, 2014a, 2014b).Within VNG, these interpretations characterize relational aspects of the semantic component which correspond in parallel to the entrenched patterns of the narrative grammar.

Narrative generativity
The combination of e-conjunction and refiners creates several options for sequencing.First, refiners ("a") could proximally follow their heads ("A") then shift to another character of the scene ("B") to form an A-a-B pattern.Refiners and heads would thus form a constituent of a single character within narrative grammar, which would then unite with the other character in e-conjunction, as in Fig. 2a.This grouping structure would uphold a semantic hierarchy positing that characters belong within scenes which then shift across time changes (i.e., characters < scenes < time), as posited in various approaches to visual narrative (Bateman & Wildfeuer, 2014a;Cohn, 2010;Magliano & Zacks, 2011).A proximal relationship between refiners and heads can apply to both anaphoric (A-a) and cataphoric (a-A) refiners (Fig. 2a and 3b), with the environment (A-B) conjoined at a higher level of structure: Another possibility is to separate a refiner from its head.In refiner displacement, a refiner follows the B-panel (Charlie in Fig. 2c and Fig. 1), resulting in an A-B-a pattern. 1 Cataphoric refiners can also be displaced, resulting in an a-B-A pattern (Fig. 2d).Refiner displacement thus introduces a distance dependency to the sequence, as the refiner must now be connected to its non-adjacent head.

Table 1
Structural schema in narrative grammar.
(a) Canonical narrative schema: 1 This structure interestingly creates an inversion of the canonical structure of anaphora in sentences.In refiner displacement, the refiner belongs to a mother node that "commands" the conjunction constituent containing its antecedent.This is the inverse structure of how anaphora within sentences operate, whereby the antecedent commands the anaphor (Chomsky, 1981).We should note that this resulting structure is not "by design," but is simply the byproduct of what emerges from applying the narrative grammar diagnostic tests to reveal the structure (Cohn, 2015).
N. Cohn et al.Cataphoric distant refiners also alter the embedding, from a left-branching structure to a right-branching structure.Right-branching can also be maintained with proximal refiners, both anaphoric (Fig. 2e) and cataphoric (Fig. 2f) so long as the B-character precedes the A-character in the sequence.Across these six variations, VNG sometimes retains the semantic constraints of characters (plus refiners) being subordinated to environments, yet does not in distance dependencies (Fig. 2c and Fig. 2d).Distant refiners are of primary interest because they flout Fig. 2. Variations of a visual narrative sequence using Environmental-Conjunction (subscript "e") and refiners.The placement of the refiner relative to its head (double bar lines) cuts across dimensions of order (anaphoric, cataphoric), distance (proximal, distant), and embedding (left-branching, right-branching).
both the expected semantic hierarchy and involve distance dependencies.Insofar as refiner displacement involves the back and forth between referential entities, it has similarities with a "shot/reverse-shot" pattern from film which alternates between views of characters in a filmed sequence (Bateman & Schmidt, 2012).Some work has examined the spatial continuities constraining attention throughout shot/reverse-shot sequences (Kraft, Cantor, & Gottdiener, 1991;Smith, 2012), and naïve film viewers can experience difficulty construing this spatial inference (Schwan & Ildirar, 2010).Yet, while the shot/reverse-shot pattern can include both characters within the same frame, only alternating which character is facing the viewer (Bordwell & Thompson, 1997), refiner displacement must individuate characters in frames, followed by a zoom of the first character.Thus, refiner displacement may be a more restricted, sister schema to the shot/reverse-shot.
While various approaches have subcategorized types of visual narrative patterns (Bateman & Schmidt, 2012;Branigan, 1992;Cohn, 2019;Metz, 1974), little empirical work has examined the constraints on such patterns in sequential images.One reason may be the widespread focus on characterizing the semantic coherence relations between images (Bateman & Wildfeuer, 2014a, 2014b;Magliano & Zacks, 2011;McCloud, 1993;Smith, 2012), and on the assumption that any sequencing of images may be construable (Bateman & Wildfeuer, 2014a, 2014b).In Fig. 2, all sequences indeed could be comprehensible and/or well-formed.However, this does not rule out the possibility that comprehenders will have preferences for how those elements should optimally pattern.Thus, our first question is: are there constraints for how these elements may be presented in a sequence?
In addition, while no research has yet empirically examined refiners in visual narratives, anaphora in language has received considerable focus (reviewed below).If indeed refiners operate in ways analogous to linguistic anaphora, we might expect similar constraints to persist in the visual modality, including those posited in discourse research (Asher & Lascarides, 2003).Obviously, the representational qualities of verbal language and visual narratives differ-the representations used in sentences (like pronouns) clearly differ from those used in visual narratives (like refiners).However, the analogy we posit between the two modalities targets a more abstract architectural level with constructs like grammatical categories, constituents, and their relational constraints, and/or the (neuro)cognitive correlates of processing such structural patterns, akin to structural and cognitive comparisons between language and music (Koelsch et al., 2005;Patel, 2003).
We thus might situate the possibilities of cross-modal similarities at Marr's (1982) levels of analysis, which distinguish between the representations involved in cognitive phenomena (computational level), the mechanisms used to compute these representations (algorithmic level), and the neural tissue involved in these mechanisms (implementational level).A computational level comparison could be that anaphora and refiners simply share characteristics within grammatical structures as analogous indexical dependencies-i.e., they both establish co-referential relationships between a less informative element and a more semantically rich antecedent.We take up this hypothesis in Experiment 1, where we explore the constraints on refiner/e-conjunction interactions across characteristics like the order and distance of a refiner and its antecedent.A stronger hypothesis would posit that these architectural analogies manifest in similar (neuro)cognitive processing between refiners and anaphora in language, in line with Marr's algorithmic or implementational levels.Indeed, similar electrophysiological responses appear to manipulations of visual narrative grammar as to syntactic structure (Cohn, 2020b), and both verbal and visual working memory systems have been implicated in visual narrative inferencing (Magliano et al., 2015).We explore this hypothesis for behavioral responses in Experiment 2 (self-paced viewing times) and neurocognition in Experiment 3 (event-related brain potentials).

Experiment 1: Constraints on refiners
Our first experiment examined participants' preferences for refiner placement in a sequence when it interacts with e-conjunction.We presented participants with sequences using e-conjunction where positions between panels were marked before or after the A-or Bpanels, as in Fig. 3. Below the sequence was a zoom panel of the A-panel, and participants were asked which open position they would place it into.We measured both where they placed the zooms into the sequence, and their response times for making this decision.Participants' choices could yield the six possible structures in Fig. 2 depending on the distance of the refiner to its head (proximal or distant), its order (cataphoric or anaphoric), and/or its resulting embedding (left-or right-branching structures).
On the basis of semantic relationships, proximal refiners should be preferred over distant ones, implying that participants prefer structures like Fig. 2a, 2b, 2e, or 2f over those in Fig. 2c or 2d.We might expect that the features of a single character-like a refiner and its antecedent-will be preferred to be kept together, to then group with other characters within a shared environment (Bateman & Wildfeuer, 2014a;Cohn, 2010).
The contrast between anaphoric and cataphoric refiners has fewer clear precedents in visual narrative research, as no work has previously examined the ordering of zoom panels relative to their antecedents.However, if our analogy with language holds true, we might expect comparable constraints to refiners as anaphora in language.There, cataphoric relations have been shown to incur greater costs than anaphoric relations (Brasoveanu & Dotlačil, 2015;Filik & Sanford, 2008;Filik, Sanford, & Leuthold, 2008;Kazanina, Lau, Lieberman, Yoshida, & Phillips, 2007).Linguistic research also supports that distance dependencies evince more costs than more proximal relations (Gibson, 2000) and a preference for dependency locality has been posited as a universal of syntactic structure (Futrell, Levy, & Gibson, 2020).
Similar expectations come from discourse theories, where anaphoric relations are said to follow a right frontier constraint (Asher, 2008;Asher & Lascarides, 2003).This constraint posits that anaphoric information is less accessible to a closed segment-i.e., attaching to a constituent following an antecedent (rightward) rather than preceding it or at a distance.Under this interpretation, a displaced refiner of [A-B-a] would lose its elaboration (A-elaboration-a) relation to its A-antecedent because it would be unable to attach on the right frontier (i.e., the B panel), and instead maintain a flat structure of three contrast relations, e.g., A-contrast-B-contrasta (Bateman & Wildfeuer, 2014a).This would differ from (or persist in the parallel semantic structures with) the structural VNG N. Cohn et al. interpretation which would maintain the distance dependency of the refiner, as in Figs. 1 and 2c.
Nevertheless, corpus research suggests a more complicated situation (Cohn, 2019).An analysis of 10,521 panels across 90 comics from several countries (United States, Sweden, France, Germany, Japan, Korea, Hong Kong) found that e-conjunction, proximal refiners, and refiner displacement occur more frequently in Asian comics than in Western comics.In addition, refiner displacement actually appears more frequently than proximal refiners in nearly all populations.Thus, if participants are familiar with these as narrative constructions through their background experience with comics (Cohn, 2020a;Cohn & Kutas, 2017), then distant anaphoric refiners (Fig. 2c) might be preferable to proximal ones (Fig. 2a), despite the less straightforward mapping to semantics.
Thus, following these linguistic constraints, we predict overall that refiners that follow their antecedents (anaphoric, cf.Fig. 2a, c, e) will be preferred to those that precede them (cataphoric, cf.Fig. 2b, d, f), while refiners that are proximal to their heads (cf.Fig. 2a, b, e, f) will be preferred to those that are distant (cf.Fig. 2c, d).In addition, we predict an interaction between order and distance such that proximal cataphoric refiners (cf.Fig. 2b) will be less preferred than distant anaphoric refiners (i.e., refiner displacement, cf.Fig. 2c).

Stimuli
We designed our stimuli using wordless panels culled from The Complete Peanuts volumes 1-12 (1950-1974) by Charles Schulz, specifically using a subset of 24 sequences used in our prior research (Cohn & Kutas, 2017).All sequences had the same basic structure, starting with an Establisher panel that sets up the characters in the situation.Two Initial panels then show each character in their own panel (characters "A" and "B"), as in Fig. 2b, before proceeding to a Peak with the sequence's climax and a subsequent Release.We used both possible orders of the A and B Initial panels (A-B, B-A), to create the left-branching and right-branching structures.
Beneath the presentation of each sequence, we displayed a zoom panel that was a refiner of the "A" character ("a").Refiner panels were created by selecting a salient sub-portion of the A or B panels which was enlarged to the size of a whole panel.Participants were tasked with deciding where to place this zoom panel in the sequence between two options of positions relative to the A and B Initial panels, as in Fig. 3. Depending on where the refiner would be placed in the sequence, these paired choices thus balance the creation of sequences that vary in their embedding (left-or right-branching), the distance of refiner relative to its antecedent (proximal, distant) and/or their serial order (cataphoric, anaphoric).All possible options, and their resulting sequence types, are outlined in Table 2.
To add heterogeneity to the experiment, fillers had refiner panels to be placed that were incongruent characters in the sequence.These panels were culled from refiners that did not appear in the experimental sequences.We distributed our 24 sequences into 12 lists counterbalanced in a Latin Square Design such that each list presented each sequence only once, with all possible pairwise choices for refiner positions appearing across lists.

Participants
We recruited 149 participants (57 male, 85 female, 7 other; mean age: 27.1) through social media and the human subjects pool of the Tilburg Humanities and Digital Sciences of Tilburg University.All participants gave their informed written consent according to approval from the Tilburg University School of Humanities and Digital Sciences Research Ethics and Data Management Committee.Participants were asked to choose a location to place the zoom panel in the sequence.This example used an "_AB_" choice that would result in either a cataphoric proximal refiner (1, cf.Fig. 2b) or an anaphoric distant refiner (2, cf.Fig. 2c), both of which would be left-branching.
Participants' familiarity with comics was measured using the Visual Language Fluency Index (VLFI) questionnaire which asks about habits of reading and drawing comics, both while growing up and currently.The VLFI score calculated from this questionnaire has consistently correlated with both behavioral and neurocognitive measures (for review, see Coderre & Cohn, 2023;Cohn, 2020a).Participants averaged a VLFI score of 11.52 (SD = 8.60, range: 1.5-45.5),which is an average proficiency (low: < 8, average: ~12, high: > 20).

Procedure
Participants took the experiment via an online interface of Qualtrics using the jspsych plugin ( de Leeuw, 2015).After screens where they could give their informed consent and fill out the VLFI questionnaire, the 24 stimulus strips were presented one at a time in an order randomized for each participant.Sequences were shown with numbered arrows pointing to the spaces beneath two of the gaps between panels.Below this sequence was a zoom panel with the instructions "Where would you place this 'zoom panel' in the sequence above?"Participants were instructed to press the number of the position where they would choose to place the zoom panel into the sequence (as in Fig. 3).At the end of the experiment, participants were asked to provide any additional commentary or insights they may have had, and then a final screen debriefed them about the intent of the study.

Data analysis
Participant choices were analyzed with a generalized linear mixed model (GLMM, package lme4 in R), predicting the binomial outcome for each position (whether it was chosen or not).We entered fixed factors of order (anaphoric or cataphoric), distance and embedding (left-branching proximal, right-branching proximal or distant) and their interaction and included random effects of participants and strips and random slopes where possible.Factor contrasts used treatment coding unless otherwise noted.Model fit was evaluated by comparing models with and without an effect of interest using Maximum Likelihood.Follow-up comparisons were carried out using the emmeans package.Our model comparisons for all analyses are provided in the Supplementary Materials.
The analysis of response times aimed at finding out whether choices for some types of contrasts were harder to make than others.Outliers, i.e., response times that were too fast (RT < 1,000 ms) or too slow (RT > 28,000 ms), were removed.This upper threshold was determined as the mean plus 2.5 times the standard deviation, which was rounded off to 28,000 ms.Valid RTs were then analysed using a linear mixed effects model, applied in the same way as for the choice analysis.P values associated with model estimates are derived using the lmerTest package.
Finally, we investigated any potential role of expertise by adding participants' VLFI scores as a fixed factor to the previous models (along with its interactions with order and distance-embedding).Because refiner constructions appear more prevalently in Asian comics than American or European comics (Cohn, 2019), we followed significant VLFI correlations with analyses of VLFI subscores related to prevalence of reading manga currently or while growing up.

Forced choice results
Refiners placed into anaphoric proximal patterns were chosen most often (see the quantity of choices depicted by dots in Fig. 4, for example the comparison between LAP and LCP).Cataphoric patterns were chosen less often.We compared the proportions of choices across narrative patterns.The best fitting GLMM (see Supplementary Material for details) indicated a significant effect of order (β = 0.74, ± 0.13 SE, p <.001), with anaphoric positions more likely to be chosen than cataphoric positions.There were also interactions between order and the levels of distance-embedding (βs > 1.04, ps < 0.001), which we inspected further with separate models for anaphoric and cataphoric orders.Within anaphoric orders, proximal orders were chosen the most, with right-branching proximal patterns chosen the most, followed by left-branching proximal, and then anaphoric distant refiners (all pairwise comparisons were reliable, p <.01).Within cataphoric refiners, distance-embedding did not improve model fit and there were no differences between levels (ps > 0.3).

Response times
We next analyzed how long it took participants to make their placement decisions, as depicted by the dots and error bars in Fig. 4. Order and distance embedding were not significant predictors of RT but there was a significant interaction (comparing distant and left-

Table 2
Patterns used in the forced choice test for where to place a refiner between panels with two characters (A and B) such that the refiner relates to the A panel.The "_" represent the open slots of the choices (in order of the sequence types).Italics indicate the variable constraint tested by each choice.branching proximal positions, β = -2200, ± 579 SE, p <.001).Within anaphoric patterns, distant positions were chosen significantly slower than left-branching proximal and right-branching proximal positions (both ps < 0.005).Panels chosen for cataphoric positions were responded to more slowly in general, but there were no reliable differences between left-and right-branching proximal positions and distant positions within this order (ps > 0.18).

Visual language fluency index
Finally, we analyzed whether comic reading experience, as measured through VLFI scores, had a relationship to participants choices or response times.The model of pattern choices was improved by adding individuals' VLFI scores as a fixed predictor, along with interaction terms (χ2(6) = 18.6, p =.005).We inspected this further with exploratory correlations by subject to investigate which choices were associated with expertise.VLFI scores correlated negatively with right-branching proximal anaphoric choices, r(1 4 7) = -0.219,p <.001, suggesting that participants with greater fluency were less likely to place refiners into positions that created this pattern.Follow up analyses again found negative correlations between this same pattern and participants' rates of reading manga while growing up, r(1 4 7) = -0.190,p <.05, and reading manga currently, r(1 4 7) = -0.201,p <.05.Adding VLFI scores did not improve the model predicting response times.

Discussion
This experiment examined participants' preferences for refiners' positioning relative to their antecedents when interacting with panels of two different characters.In line with our predictions, participants preferred to place refiners into positions where they followed their antecedents (anaphoric) rather than preceded them (cataphoric), and to be proximal to their antecedents rather than distant.These preferences mostly follow expectations for the basic semantic constraints for the construction of characters, scenes, and then temporal relations (Bateman & Wildfeuer, 2014a;Cohn, 2010;Magliano & Zacks, 2011), for discourse relations (Asher & Lascarides, 2003;Bateman & Wildfeuer, 2014a), and for the relative constraints of anaphoric relations in sentence processing (Brasoveanu & Dotlačil, 2015;Filik & Sanford, 2008;Filik et al., 2008;Kazanina et al., 2007).However, semantic constraints alone cannot explain the full pattern of our results.Refiner order appeared to be the strongest predictor of preferences.Participants by far chose to create anaphoric, rather than cataphoric refiners.This effect was most pronounced when the proximal position was held constant, regardless of the embedding: across choices, anaphoric refiners placed into a proximal position were chosen 76 % of the time compared to a cataphoric proximal position (23 %), and these choices also maintained the fastest response times.This preference had less to do with the semantic construction of scenes (characters > scenes > time), but rather with the placement of focal information relative to a full-scene.Placing refiners after their antecedents implies that comprehenders prefer to view more information first and then focal details, rather than starting with partial information and then expanding on it.This preference was strongest for right-branching anaphoric refiners (81 %) where the "preferred" semantic sequence is inverted, but is the most adhering to a constraint of the right frontier (i.e., the preferred [B-[A-a]] sequence constructs the scene out of B-and A-panels, then adds the refiner "a", which would be the opposite if the scene was built from smaller to larger information with an [[a-A]-B] sequence).
We also observed constraints of distance, but only within the anaphoric refiners, as cataphoric positions did not differ in the proportion that they were chosen.Preferences for placing refiners into anaphoric distant positions overall were between the highly preferred (anaphoric proximal) and dispreferred orders (cataphoric).Indeed, despite an overall preference for proximality and these sequences' violation of a right frontier constraint, anaphoric distant refiner positions were chosen a substantial amount when compared against anaphoric proximal (36 %) and cataphoric proximal (38 %) positions, which was closer than the (80/20 %) differential observed in other comparisons.In addition, choices related to anaphoric distant refiner positions evoked the longest response times, suggesting that participants required more time to consider whether these patterns were well-formed.The longer response times perhaps reflects a tension between their violation of semantic constraints with their status as narrative constructions.
Overall, as predicted, both order (anaphoric > cataphoric) and distance (proximal > distant) constrain participants' intuitions for visual narrative sequencing.Tensions between these constraints were evident in anaphoric distant (LAD) or cataphoric proximal (LCP) left-branching refiner positions evoking much slower response times than the canonical anaphoric proximal positions (LAP).Each of these patterns maintains one preferred constraint while violating another.Follow up analyses revealed comparable response times when these positions were directly compared (LAD vs. LCP:_AB_), where participants ultimately placed refiners into distant anaphoric positions (refiner displacement: A-B-a) more often (61 %) than cataphoric proximal positions (a-A-B).These similar response times suggest that participants had a harder time weighing these constraining factors when pitted against each other, where anaphoric relations eventually won out.
Finally, right-branching anaphoric proximal positions were the only pattern to correlate with participants' comic reading expertise (VLFI scores), particularly manga reading.These correlations suggest that increased fluency leads to reduced preference for this pattern-perhaps due to its infrequency within actual manga-despite being chosen the most of all structural patterns overall (81 %).This again may reflect a tension between semantic preferences (like the right frontier) and recognition of its status as a narrative construction.

Experiment 2: Self-paced viewing
Experiment 1 suggested that participants had two main preferences for the location of refiners relative to their antecedents: order (anaphoric > cataphoric) and distance (proximal > distant).Such preferences contrast with claims that the structure of such sequences is relatively unconstrained (Bateman & Wildfeuer, 2014a), and they align with general preferences of anaphoric relations in language (Brasoveanu & Dotlačil, 2015;Filik & Sanford, 2008;Filik et al., 2008;Kazanina et al., 2007).We next asked whether order and distance would modulate participants' online comprehension.We therefore measured participants' self-paced viewing times for each panel in visual sequences that modulated both the distance (proximal, distant) and order (anaphoric, cataphoric) of refiners.
One possibility is that refiners will be viewed faster than their heads simply because they are zoom panels depicting less information.Refiners by definition show a constrained amount of information relative to a full depiction of a single character.This focal information should require less attentional selection and information extraction processes, thus facilitating subsequent comprehension processes (Loschky et al., 2020).Nevertheless, self-paced viewing times did not differ when comparing full-scenes and zoomed-in panels in the same position of a sequence (Foulsham & Cohn, 2021).Yet, refiners do not just provide a zoomed-in view as an alternative to a full-scene panel, but they repeat the information in their antecedent.Because of this, the reactivation of information should facilitate processing, as the repetition would require less updating into a mental model (Cohn, 2020b;Loschky et al., 2020).This would imply that ordinal sequence position will influence processing such that whichever element comes second-refiner or head-will be processed more easily than whichever comes first.Yet, Experiment 1 implied a dispreference for refiners preceding their heads.Thus, while order may affect processing, reactivation of less information (anaphoric refiners) may be easier to process than reactivation plus more information (antecedents following cataphoric refiners).Such costs would be consistent with longer self-paced reading times shown to cataphoric relationships in sentence processing (Brasoveanu & Dotlačil, 2015;Drummer & Felser, 2018;Kazanina et al., 2007;Kazanina & Phillips, 2010) and with the idea that richer information in encoding leads to facilitation in subsequent retrieval (Hofmeister, 2011).
Following the results of Experiment 1, distance dependencies were predicted to be harder to process than proximal relationships.In anaphoric distance dependencies, the repetition of the zoomed information may facilitate reactivation, but increased distance may N. Cohn et al. ameliorate such effects because of the intervening B-character.This would be consistent with a discourse interpretation following a right frontier constraint, where the anaphoric distant refiner would yield only a local contrast relation (A-contrast-B-contrast-a), and thus we might expect no effect of the distant connection between head (A) and refiner (a) that would be maintained in a VNG interpretation.Indeed, while no studies of visual narrative sequencing have yet manipulated distance dependencies outright, longer reading times occur to distance dependencies in sentence processing (Gibson, 2000;Gibson & Warren, 2004;Hofmeister & Sag, 2010).
Finally, processing differences may carry over to panels following the Initial constituent.Prior work has shown that individuating characters in scenes using e-conjunction facilitates predictions of subsequent events (Kaiser, 2019), which can be modulated by the order of the conjoined panels (Cohn & Paczynski, 2013).In addition, disruptions to narrative constituent structure have been shown to affect the processing of panels further downstream in the sequence (Cohn, Holcomb, Jackendoff, & Kuperberg, 2012).Indeed, distance dependencies and anaphora in sentence processing often manifest costs downstream from the critical words themselves (Badecker & Straub, 2002;Brasoveanu & Dotlačil, 2015;Kazanina et al., 2007).Thus, we also examined viewing times for panels subsequent to the manipulation to assess any spillover of these narrative constructions.Here we again predicted that such downstream costs will be greater for panels following cataphoric and/or distant refiner relationships.

Stimuli
In Experiment 2, we used the same sequences as in Experiment 1, though now only those that have the refiners fully integrated into the sequences.Each strip began with an Establisher that introduced both characters.The subsequent three panels showed the refinerconjunction interactions (positions 2, 3, 4), after which the final two panels (5, 6) depicted a climactic Peak and a Release.In the critical panel positions 2-4, we balanced the variables of the serial order (cataphoric, anaphoric) and distance (proximal, distant) of refiners relative to their head.This resulted in four conditions of sequence types, consistent with the patterns in Fig. 2a-d (i.e., A-a-B, A-B-a, a-A-B, a-B-A).
To prevent confounds of any agent-patient or left-right relationships between conjunction panels (Cohn & Paczynski, 2013), both orders of A-B and B-A panels were used, along with refiners for both the A and B panels for their appropriate sequences (e.g., anaphoric proximal: A-a-B and B-b-A).Our final analysis collapsed across A-B and B-A sequence orders.An additional 8 filler sequences included incoherent panels to introduce heterogeneity into the experimental stimuli.These sequence types were distributed in eight lists with 24 experimental sequences and 8 fillers each, counterbalanced using a Latin Square Design such that each sequence appeared only once per list with even numbers of each sequence type, but all sequences appeared in all experimental conditions across lists.Each participant viewed a uniquely ordered, randomized list.

Participants
We recruited 161 participants (70 male, 87 female, and 4 other, mean age: 28.2) using the same methods as in Experiment 1. Their mean VLFI score was average, at 12.38 (SD = 9.67).

Procedure
As in Experiment 1, participants took the experiment online via Qualtrics using the jspsych plugin ( de Leeuw, 2015).After giving informed consent and filling in the VLFI questionnaire, they were presented with each trial sequence one panel at a time.Following a screen reading "Ready", participants advanced through each panel in a sequence at their own pace by pressing a button.At the end of each sequence, a question mark appeared on the screen along with a 1 to 7 scale asking participants to rate the comprehensibility of the strip on a scale of 1 (very difficult) to 7 (very easy), by pressing the number on the keyboard.All 24 sequences for each participant were presented in a randomized order.Following the experimental sequences, participants were given a space to share any thoughts or observations about the study, and then were subsequently debriefed.

Data analysis
Our analysis focused on the self-paced viewing times of the three critical panels (positions 2, 3, and 4) and the comprehensibility ratings for the whole sequences.These were analyzed using linear mixed models with fixed factors of order (cataphoric, anaphoric) and distance (proximal, distant) and random effects of participant and strip.Random slopes were included where possible.Our analysis of viewing times first used an outlier removal process that excluded times that were too fast (<300 ms) or too slow (>3028.53ms), using the same outlier removal methods as in Experiment 1.Additional models including VLFI scores were also examined.

Ratings and post-test questionnaires
A model with comprehensibility rating as the outcome found no reliable prediction from order or distance, or their interaction (vs. a null model, χ2(3) = 1.78, p =.62).Ratings for anaphoric refiners did not differ between proximal (M = 5.45, SD = 0.97) and distant locations (M = 5.43, SD = 0.93) while cataphoric refiners also did not differ between proximal (M = 5.47, SD = 0.96) and distant locations (M = 5.38, SD = 0.98).On average, all sequences were deemed as above average (4) in the 1 to 7 scale of comprehensibility.
In post-test questionnaires, only 21 % of the participants (23/112) noticed without prompting the "close-ups" or "zooms," while 16 % (18/112) mentioned the inverted order of Initial panels in the B-A sequences.No participants mentioned the distance of refiners relative to their antecedents.
N. Cohn et al.

Self-paced viewing times
Self-paced viewing times across all 6 panel positions for the sequence types are depicted in Fig. 5. Overall, viewing times were longest at the first panel, grew faster at the critical panel positions 2, 3, and 4, and then slowed again for panels 5 and 6.Refiner panels (squares in Fig. 5) were noticeably faster in anaphoric relationships (black lines) compared to cataphoric relationships (red squares, at panel position 2).At panel position 5, we also observed the effects of both factors, with the longest times to panels following anaphoric distant refiners, intermediate times to those after cataphoric refiners, and the fastest times to those after anaphoric proximal refiners.
Our statistical analyses focused on the three critical positions 2, 3, and 4, which depicted the conjoined characters (A, B) and refiner panels.For this analysis we used sum contrasts for the factors of order and distance to enable easier interpretation of interactions.Panel type was a significant predictor, with refiners being viewed shorter than the other types of panels (vs.antecedents, β = 69.8,± 8.7 SE, p <.001; vs. other character, β = 60.8, ± 8.7 SE, p <.001).This can be seen in Fig. 5 where the refiner panels are visibly faster than the other panels at panel positions 2, 3, and 4. A main effect of refiner order reflected that overall, panels in cataphoric sequences were viewed slower than those in anaphoric sequences, (β = 36.4,± 6.2 SE, p <.001).Refiner distance was not a reliable predictor (β = -3.43,± 6.2 SE, p =.58).The best fitting model also included significant interactions between panel type and order, but interactions involving distance were not reliable.The interaction between panel type and order emerged because anaphoric refiners were viewed faster than cataphoric refiners (p <.001), whereas there was no difference between viewing times of A and B panels in different order conditions (both ps > 0.4).In Fig. 5 this is visible in that the refiners at panel positions 3 and 4 (black squares) are faster than the refiners at position 2 (red squares).
Additional analyses focused on the non-critical panels.We observed no difference between sequence types at the first or final panel positions (predictors provide no improvement beyond a null model, both χ2(3) < 1.5, p >.68).At panel 5-the first position subsequent to the critical panels-the best fitting model included factors of order, distance and their interaction.The interaction between distance and order, β = -25.3,± 8.1 SE, p =.002, clarified that panels following anaphoric distant refiners were slower than panels following anaphoric proximal refiners (p <.001).Cataphoric refiners had intermediate viewing times that did not differ from other sequence types.

Visual language fluency index
Adding VLFI scores to the above models as a continuous predictor improved models of both the ratings for particular sequences (χ2 (1) = 4.2, p =.04) and the viewing times for all panels (χ2(1) = 22.5, p <.001).We again explored these with correlations at the subject level, to examine whether more expert readers gave different ratings or had different viewing times, and in which conditions.Positive correlations between VLFI scores and the ratings for both anaphoric proximal, r(1 5 9) = 0.17, p <.05, and anaphoric distant sequences, r(1 5 9) = 0.22, p <.01, suggested that greater fluency led to higher judgements of comprehensibility.VLFI scores also negatively Fig. 5. Mean viewing times per panel across sequence types which crossed the distance of refiners relative to its antecedent (proximal, distant) and their serial order (cataphoric, anaphoric).Markers designate refiners (square), antecedents "A" (diamond) and the additional "B" character (triangle).
N. Cohn et al. correlated with the difference between ratings for cataphoric and anaphoric sequences, r(1 5 9) = -0.176,p <.05, implying that greater fluency reduced the difference in their comprehensibility.Viewing times for all panels in all sequences positively correlated with VLFI scores, such that participants with greater fluency viewed panels faster.Correlations were weakest at first panels of all sequences (all rs > 0.195, all ps < 0.05), and stronger for subsequent panels (all rs > 0.227, all ps < 0.001).

Discussion
This experiment examined the online processing of distance and order of refiners relative to their antecedents using a self-paced viewing task.Overall, participants' ratings support that the sequences were equally construable (Bateman & Wildfeuer, 2014a), but modulated by participants' expertise for the more entrenched anaphoric refiner sequences.Viewing times were longer to the first panel, sped up for the Initial constituent (panels 2, 3, 4), and then slowed down again for the final two panel positions (i.e., the Peak and Release).The longer times at the first panel, and decrease at subsequent positions, aligns with prior observations of longer viewing times at the first panel of visual narrative sequences (Cohn & Paczynski, 2013;Cohn & Wittenberg, 2015;Foulsham et al., 2016).Relative differences between ordinal sequence positions may also reflect variance in the amount of information of the different panels, as full scenes (positions 1, 5, 6) were viewed slower than panels depicting single characters (A-, B-panels) and zooms of characters (refiners).
At our critical panel positions (2, 3, 4) this difference in the amount of information persisted with shorter viewing times overall to refiners and single character panels (A, B).Here, single character panels did not differ based on sequence position, but refiner panels did.Cataphoric refiners, which appeared at panel position 2, were viewed longer than both anaphoric proximal (panel position 3) and distant refiners (panel position 4).Despite viewing times typically decreasing across ordinal positions of panels (Cohn & Paczynski, 2013;Cohn & Wittenberg, 2015;Foulsham et al., 2016), no such difference of position continued between anaphoric proximal (position 3) or distant refiners (position 4).
Cataphoric refiners could have been viewed slower because zoom panels appearing early in the sequence do not yet mark them as refiners.This would align with prior work showing zoom panels not to differ in viewing times from full scenes (Foulsham & Cohn, 2021).These zoom panels may thus be unexpected, reflecting the dispreference for cataphoric refiners in Experiment 1.This would be consistent with the suspension of incremental processing found to cataphoric pronouns in sentence processing (Filik & Sanford, 2008).Conversely, rather than the cataphoric refiners being viewed relatively slower, anaphoric refiners may have become faster due to their prior antecedent.That is, cataphoric refiners reflect the baseline processing for zoom panels while presenting this information for the first time, while the information repeated in anaphoric refiners from previous antecedents may confer a processing advantage.As a result, priming from the antecedent allows anaphoric refiners to be viewed faster than cataphoric refiners.Such an interpretation would also affirm the distance dependency in anaphoric distant refiners, which would receive this facilitation despite their separation from an antecedent.This also does not follow in reverse: antecedent A-panels receive no viewing time benefit from following a cataphoric refiner.It thus seems that if preceding panels confer a priming effect, it flows from a full-scene to focal information, but not the reverse.This directionality follows the amount of information: An anaphoric refiner repeats a focal part of a previously-viewed whole character, filtering out the no-longer-relevant information.In contrast, a full-scene after a cataphoric refiner now incurs the cost of adding information to the focal zoom, while also recognizing the zoom as part of that scene.This aligns with findings that richer linguistic representations of referents can facilitate viewing times to their subsequent access (Hofmeister, 2011).
Following the Initial constituent, panels in positions 5 and 6 of the sequence were viewed longer than the previous three panels (positions 2, 3, 4).This slowing between panels 4 and 5 differs from the typical pattern of viewing times growing faster or stabilizing at each subsequent ordinal sequence position (Cohn & Wittenberg, 2015;Foulsham et al., 2016).This slowing could again be due to the amount of information of the panels at the previous positions (2,3,4) showing either single characters or zooms, relative to the mostly full scenes at positions 5 and 6.However, zoom panels do not always differ in their viewing times from full scenes (Foulsham & Cohn, 2021), suggesting that amount of information alone may not modulate viewing times.
Longer viewing times overall at position 5 could be due to exiting of the subordinated constituent at this position, which should evoke a structural revision (Cohn, 2020b).Indeed, longer self-paced viewing times occur to panels following a break in narrative constituent structure (Cohn, Holcomb, et al., 2012), and neural responses suggestive of updating processes appear to panels following disruptions of narrative constituent structure (Cohn, Jackendoff, Holcomb, & Kuperberg, 2014).
At position 5, participants viewed panels following both types of cataphoric refiner sequences (a-A-B, a-B-A) slower than panels following anaphoric proximal refiner sequences (A-a-B).However, panels following the refiner displacement (A-B-a) were viewed the slowest, suggesting that the costs of processing the anaphoric distance dependency spilled over to this subsequent panel.Indeed, this refiner was viewed for only ~ 690 ms, which was faster than all other full character panels at position 4.In both visual narrative and sentence processing, mechanisms related to both structural revision and/or further interpretation of referential dependencies start around 500 ms (Baggio, 2018;Cohn, 2020b;Coopmans & Nieuwland, 2020), and spillover effects are common to the resolution of anaphora in self-paced reading paradigms of sentence processing (Badecker & Straub, 2002;Brasoveanu & Dotlačil, 2015;Kazanina et al., 2007).Thus, the longer times at the panel subsequent to an anaphoric distant refiner may reflect spillover of sustained processing for the distance dependency of refiner to its antecedent.Together with the facilitation to the distant refiner at position 4, this subsequent spillover effect at position 5 suggests that participants indeed create the co-referential distance dependency between refiner to its antecedent (A-B-a), not just a local contrast relation (…B-contrast-a).

Experiment 3: Event-related potentials
Our first experiments established that refiners are preferred when they follow their antecedents at a proximal distance (Experiment 1), and that deviation from these constraints incurs costs evident in self-paced viewing times (Experiment 2).While such constraints align with preferences for anaphora in sentence processing, these findings do not necessarily imply similar cognitive mechanisms operating across modalities.Thus, in Experiment 3 we investigate the neurocognitive processes of these refiner relationships for the same types of sequences as in Experiment 2.Here we measure event-related brain potentials (ERPs), which provide a direct measure of the electrical activity of the brain, often evincing responses indexing functionally distinct mechanisms.Such indices offer greater sensitivity than the viewing times in Experiment 1, thereby potentially clarifying the costs incurred for distance dependencies, or differences between refiners and other panels beyond their amount of information.
Experiment 3 thus analyzes ERPs to similar sequences as in Experiment 2. In prior work, we analyzed these data to explore coreferential processing by contrasting antecedent full character panels (A panels) and their refiners at both the cataphoric (position 2) and anaphoric (position 4) locations (Coopmans & Cohn, 2022).Here, we focus our analysis on refiner positioning in line with Experiment 2: position 4, where a distance dependency would manifest, and refiner panels, where we can observe the relative effects of sequence context on each refiner.
Research measuring ERPs on visual narratives supports a similarity in the processing of anaphoric relations across modalities (Cohn, 2020b).In the study of language, distinct ERPs index the processing of semantics and syntax (Kutas et al., 2017;Sprouse & Schütze, 2019).Semantic processing is indexed by the N400, a widespread negativity peaking at around 400 ms that is thought to reflect the access and/or integration of information from semantic memory (Kutas & Federmeier, 2011), an effect that is largely insensitive to manipulations of grammar (Van Petten & Kutas, 1991).Conversely, phasic anterior negativities often with a left lateralization ("left anterior negativity", LAN) appear in a similar time window to violations of grammatical structure (Friederici, 2011;Neville, Nicol, Barss, Forster, & Garrett, 1991).These anterior negativities are thought to index the cost of structural predictions (Yano, 2018), working memory processes (Barkley, Kluender, & Kutas, 2015), or combinatorial structure building (Baggio, 2018;Hagoort, 2017).
In line with the idea that visual narratives involve analogous processing mechanisms as language, studies of visual narratives have found similar ERPs as those in sentence processing (for review, see Cohn, 2020b).Larger N400s have appeared to incongruous or unexpected information in visual sequences (Coderre, O'Donnell, O'Rourke, & Cohn, 2020;West & Holcomb, 2002), and this appears to be insensitive to narrative structure (Cohn, Paczynski, Jackendoff, Holcomb, & Kuperberg, 2012).Unlike in language though, the N400 to images often are preceded by an N300, which is thought to index a process of object categorization or identification, or structural mapping of visual features to semantic memory (Draschkow, Heikel, Võ, Fiebach, & Sassenhagen, 2018;Hamm, Johnson, & Kirk, 2002;Schendan & Kutas, 2003;Truman & Mudrik, 2018).Indeed, zoom panels elicit attenuated N300s compared to full scenes (Cohn & Foulsham, 2020), and prior analyses of the ERP dataset described below has already shown that refiner panels evoke such a reduced N300 compared to their corresponding head (A) panels, seemingly with no subsequent N400 effect (Coopmans & Cohn, 2022).
In addition, sustained anterior negativities appear to coreferential processing, reminiscent of the Nref to anaphoric relations in language (Baggio, 2018;Coopmans & Nieuwland, 2020;Hoeks & Brouwer, 2014;van Berkum, 2012).In prior analyses of this dataset, such sustained negativities were shown to anaphoric distant refiners and, with a slightly later onset, to the antecedents of cataphoric distant refiners, both relative to their preceding coreferential units (Coopmans & Cohn, 2022).This suggested that such negativities indexed processes of coreference, regardless of whether they showed a zoom or full scene.
Unlike prior analyses which focused on coreference of refiners and their antecedents specifically, we here examine the interaction between order and distance on the processing of narrative structures.Specifically, we examined the processing of the distance dependency created by anaphoric distant refiners relative to those with proximal and/or cataphoric positions.We therefore analyzed panels at the fourth position of the sequence-the location of the anaphoric distant refiners-and compared those across the four types of refiner panels themselves.
Given prior analyses of these data, we expect that anaphoric proximal refiners show a similar sustained negativity (Nref) as to anaphoric distant refiners, reflecting the coreferential processes evoked at these panels.This should contrast with the cataphoric refiners which have yet to evoke such coreference.Nevertheless, preceding this sustained negativity, we first expect that anaphoric refiners of both distances lead to attenuated N400s, since the repetition of content should reactivate this information, as for anaphoric repeated names in discourse (Coopmans & Nieuwland, 2020).Cataphoric refiners would thus receive no such benefit, preceding their antecedents, and may even evoke larger N400s with their introduction of new information (Filik et al., 2008).Another possibility is that the N400 to refiners is modulated only by sequence position, leading to larger N400s to cataphoric refiners (position 2), then anaphoric proximal refiners (position 3), and then anaphoric distant refiners (position 4).We additionally examined the effect of the distance dependencies created by the anaphoric distant refiners and the cataphoric distant antecedents.Following the sentence processing literature, the distance dependencies may also be expected to evoke such a sustained negativity (Kutas et al., 2017;Kwon et al., 2013;Phillips et al., 2005), and both anaphora and distance dependencies have also been shown to evoke left anterior negativities (Anderson & Holcomb, 2005;Barkley et al., 2015;Kwon et al., 2013).These precedents suggest that the refiners or antecedents involved in distance dependencies will also evoke sustained and/or lateralized anterior negativities.

Stimuli
We expanded the experimental stimuli in Experiment 2 to 96 total base sequences.Our contrasts again crossed refiner distance (proximal, distant) and order (cataphoric, anaphoric), as in Fig. 2a-d.Our final analysis collapsed across both A-B and B-A orders.We also included 72 filler sequences, half with zoom panels that were not involved in refiner-head relationships.Fillers included both congruent and incongruent zooms (i.e., 25 % congruent zooms, 25 % incongruent zooms, 25 % congruent non-zooms, 25 % incongruent non-zooms).These sequences were included to increase the heterogeneity of the stimuli, so that not all zoom panels would be expected to be involved in refiner relationships, such as cataphoric refiners where a subsequent antecedent may otherwise be expected.All sequences were counterbalanced into 8 lists in a Latin Square Design such that each participant viewed each sequence once, but across participants all variations in sequence types were viewed.Each list thus had 24 sequences of each sequence type (4 x 24 = 96) along with 72 fillers, for a total of 168 sequences, which were presented in a unique randomized order per participant using PsychoPy2 (Peirce et al., 2019).

Participants
We analyzed the results from 32 participants from Tilburg University (16 male, 16 female; mean age: 22.3, SD = 3.6).Data from one additional participant was rejected due to excessive artifact from eye movement and muscle tension (>10 % of trials rejected).Participants were right-handed with no history of head trauma, taking no psychiatric medications, and with normal or corrected-tonormal vision.They had a mean VLFI score of 14 (SD = 6.6, range: 2.9-34).

Procedure
Participants sat in a chair in a soundproof chamber across from a computer screen.We kept the lights off, except for backlighting behind the screen which served to prevent a flashing effect of the stimuli which could induce blinks.Stimuli were presented using PsychoPy2 (Peirce et al., 2019).At the start of each trial, participants viewed a grey screen reading "Ready" in white letters, which also had a red dot in the center of the screen that persisted as a fixation point across all screens.After pressing a button to start the sequence, each panel appeared one at a time in the center of the screen for a duration of 1350 ms separated by a 300 ms ISI.Each black and white panel appeared on the grey screen at a size of 10.16 x 8.04 cm, yielding a horizontal visual angle of 5.2 • and vertical of 4.2 • .After each sequence a question mark appeared where participants rated the comprehensibility of the sequence using a 1 (=hard to understand) to 7 (=easy to understand) Likert scale.After the experimental trials, participants filled out a post-test questionnaire where they could share observations of patterns or "unusual" stimuli with open-response answers.

Data analysis
We measured EEG with a Brain Products ActiChamp and 32 channel Standard actiCAPs, referenced online to electrode Fz, with eye movements and blinks monitored with electrodes beside the left eye and beneath the right eye.Our sampling rate was set to 250 Hz with a high cutoff filter of 70 Hz.Electrode impedances were kept below 10 kΩ for all electrodes.Using the ERPLAB plugin for EEGLAB in MATLAB (Lopez-Calderon & Luck, 2014), we refiltered EEG data offline with a bandpass filter of 0.1-30 Hz, and re-referenced to the average of the mastoid channels (TP9, TP10).We isolated and removed trials with excessive blink or muscle artifact using ICA.We excluded participants who exceeded a rejection rate of over 10 % of trials, which resulted in leaving out one participant from our final analysis.Our analysis averaged across epochs of 1500 ms after the start of each critical panel, along with a 200 ms prestimulus baseline correction.
We focused our analysis of ERPs on the critical panel of position 4 and between the various refiner panels in the epochs of 200-300 ms, 300-500 ms, 500-800 ms, and 800-1100 ms, corresponding to ERP components of the N300, N400, and later effects (sustained negativities/Nref, P600, late frontal positivity).These epochs were chosen to be consistent with those in other studies of visual narratives (Cohn & Foulsham, 2020).
Our primary analysis used five regions of interest covering across the scalp, where we averaged across four electrodes per region.These included a central region (FC1, FC2, CP1, CP2), and peripheral regions of the left anterior (Fp1, F7, F3, FC5), right anterior (Fp2, F8, F4, FC6), left posterior (CP5, P3, P7, O1), and right anterior (CP6, P4, P8, O2).Analysis of each epoch used repeated-measures N. Cohn et al.ANOVAs with factors of Order (2: anaphoric, cataphoric) and Distance (2: proximal, distant), while peripheral regions divided the scalp into factors of Hemisphere (2: left, right) and Anterior-Posterior (AP) Distribution (2: anterior, posterior).We followed up significant interactions in the omnibus ANOVA of peripheral regions by comparing Order and Distance within each region of interest, with post-hoc comparisons using a Bonferroni correction.Comprehension ratings were compared using a repeated-measures ANOVA with factors of Distance and Order.Pearson's correlations with an alpha level set to 0.05 were used to analyze VLFI scores with ratings and the mean amplitude differences averaged across all regions for the differential amplitude for order (cataphoric minus anaphoric) or distance (distant minus proximal).

Event-related potentials
Panel 4 We first analyzed ERPs at the fourth panel position, where possible distance dependencies could occur.Distance dependencies of anaphoric distant refiners (the only refiner at this position) were contrasted with the cataphoric distant antecedents (A-panel), and the B-panels subsequent to both anaphoric and cataphoric proximal dependencies.ERPs are depicted in Fig. 6 with statistics for analyses across peripheral regions summarized in Table 3.
In the 200-300 ms epoch, Distance x AP Distribution and Distance x Hemisphere interactions were found in peripheral regions.
Analyses at each region found Order x Distance interactions in the central and both anterior regions (all F's > 4.8, all ps < 0.05, η 2 p >.134), and a main effect of Distance in the left posterior region, F(1,31) = 6.1, p <.05, η 2 p =.166.These results arose because of an attenuated negativity (N300) to the refiner (anaphoric distant) compared to the three other non-zoom panels.Yet, panels from cataphoric sequences were also less negative than those to panels in the anaphoric proximal sequences.
In the subsequent 300-500 ms epoch, a Distance x Hemisphere interaction was found in the omnibus analysis of peripheral regions.
Follow up analyses found a main effect of Order in the right anterior region, F(1,31) = 5.6, p <.05, η 2 p =.152, because a frontal negativity began around 300 ms with a peak around 400 ms to the anaphoric refiners.An additional main effect of Distance appeared in the left posterior region, F(1,31) = 9.9, p <.005, η 2 p =.242, because both distance dependencies (anaphoric refiner and cataphoric distant A panels) evoked larger leftward negativities than the panels from proximal sequences.No significant findings were found in the central, left anterior, or right posterior regions.
Both the frontal and leftward negativities sustained across later epochs, where main effects of both Order and Distance appeared in both the 500-800 ms and 800-1100 ms time windows along with an Order x Distance x Hemisphere interaction (800-1100 ms).First, a  both posterior regions (all Fs > 4.5, all ps < 0.05, η 2 p >.126).

Refiners.
We next compared the ERPs to the refiner panels, collapsing across sequence positions in each sequence type (Fig. 7).In the 200-300 ms epoch, a main effect of Distance and an Order x Hemisphere interaction appeared in the analysis of peripheral regions (Table 3).Main effects of Distance appeared at the central and both anterior regions (all Fs > 5.3, all ps < 0.05) as did Order x Distance interactions which came close but did not exceed the threshold of significance (all Fs > 3.6, all ps < 0.072), while a main effect of Order arose in the right posterior region, F(1,31) = 5.9, p <.05.These interactions imply that a widespread frontal negativity (N300) was attenuated for all refiners except anaphoric distant refiners.
In the 300-500 ms epoch, our omnibus analysis showed only an Order by Distance interaction.Analysis at each region revealed Order x Distance interactions in the central and right anterior regions (all Fs > 4.3, all ps < 0.05, η 2 p >.122), but only a trend of an interaction occurred in the left anterior region, F(1,31) = 3.1, p =.087, η 2 p =.092.Here, an attenuated frontal-central negativity (N400) was evinced by the anaphoric proximal refiners compared to anaphoric distant and both cataphoric refiners.
The Order x Distance interactions continued in the 500-800 ms epoch, which also saw Distance x Hemisphere interactions.Analysis of regions clarified that Order x Distance interactions occurred in both posterior regions (all Fs > 6.7, all ps < 0.05, η 2 p >.178).This arose because anaphoric distant refiners evinced a larger parietal negativity with a slight rightward distribution than the other refiners.An additional marginal effect that did not reach the threshold of significance was found for Order in the right anterior region, F(1,31) = 3.9, p =.055, η 2 p =.114, reflecting a greater negativity to anaphoric refiners than cataphoric refiners.Both these effects continued into the 800-1100 ms epoch, with a main effect of Order, and Order x Distance and Order x Hemisphere interactions appearing in peripheral regions.These manifested because the posterior negativity to anaphoric distant refiners continued in the right posterior region with a main effect of Distance, F(1,31) = 6.0, p <.05, η 2 p =.162.In addition, main effects of Order appeared at the central and anterior regions (all Fs > 10.1, all ps < 0.005, η 2 p >.245).These findings reflect a greater frontocentral negativity to anaphoric refiners than cataphoric refiners, while an additional interaction arose between Order and Distance in the left anterior region, F(1,31) = 4.6, p <.05, η 2 p =.130, because anaphoric distant refiners evoked an even greater negativity in this region than anaphoric proximal refiners.

Visual language fluency index
We correlated participants' VLFI scores with the average amplitude of all regions within an epoch for the differential amplitude for order (cataphoric minus anaphoric) or distance (distant minus proximal).VLFI scores negatively correlated with the effect of order in the 500-800 ms epoch, r(30) = -0.385,p <.05.This suggested that the sustained negative effect for anaphoric panels relative to cataphoric panels was reduced for participants with greater comic reading expertise.

Discussion
We furthered the comparison between refiners in visual narrative and anaphoric relations in language by examining whether refiners evoke similar neurocognitive responses as observed in language processing.We measured ERPs, which also afforded more sensitivity to processing than allowed by the self-paced viewing paradigm in Experiment 2.
First, when compared to panels of full characters at panel 4, refiner panels evinced a reduced N300.This replicates observations of attenuated N300s to zoomed-in panels compared to full scenes (Cohn & Foulsham, 2020) and prior comparison of distant refiners and their antecedents (Coopmans & Cohn, 2022).As the N300 has been interpreted as an index of object categorization or identification (Draschkow et al., 2018;Hamm et al., 2002;Schendan & Kutas, 2003;Truman & Mudrik, 2018), the attenuation effects imply that zoomed-in viewpoints require less cost for identification processes than depictions with more information.These findings are consistent with the reduced viewing times to all refiners compared with other panels in Experiment 2, interpreted as differences in the amount of information.Nevertheless, anaphoric distant refiners evoked a greater N300 than the other refiners, implying that the Bpanel intervening between the antecedent and the refiner in these sequences mitigated the reduction of the N300 compared to those in other sequences.
Comparison between refiners showed a similar non-attenuation for anaphoric distant refiners compared to attenuated anaphoric proximal refiners in the subsequent 300-500 ms epoch.Here, anaphoric distant refiners and cataphoric refiners displayed a larger centro-anterior negativity similar to an N400 than anaphoric proximal refiners.Attenuated N400s similar to the anaphoric proximal refiners have appeared for referentially dependent repeated names in studies of anaphoric relations in language (Coopmans & Nieuwland, 2020;Nieuwland, Coopmans, & Sommers, 2019), while enhanced N400s are evinced by pronouns with no preceding antecedent (Filik et al., 2008).N400s are typically attenuated across sequence position (Cohn, Paczynski, et al., 2012;Van Petten & Kutas, 1991), which would lead cataphoric refiners (position 2) to have greater amplitudes than anaphoric refiners (positions 3 and 4), as was found here.However, if this was an effect of position alone, anaphoric proximal refiners (position 3) should have a greater amplitude than anaphoric distant ones (position 4).Yet, here only anaphoric proximal refiners demonstrated an attenuation, likely driven by the immediate repetition of referential information, while anaphoric distant refiners did not differ in amplitude from cataphoric refiners.Thus, the lack of N400 attenuation to the anaphoric distant refiners, ameliorated by the intervening B-panel, suggests that their position in the sequence rendered costs beyond that of anaphoric proximal refiners.
In addition, a sustained anterior negativity in later time windows (500-1100 ms) appeared at position 4 only for anaphoric distant refiners, and between refiners for anaphoric compared to cataphoric refiners.This was consistent with the findings in language of a late sustained anterior negativity (SAN/Nref) to referential dependencies (Baggio, 2018;Hoeks & Brouwer, 2014;van Berkum, 2012), and to prior analysis of this dataset (Coopmans & Cohn, 2022).Such negativities suggest a greater cost of establishing a referential N. Cohn et al. dependency for a refiner that follows its antecedent rather than one that precedes it.Such an effect also reinforces the connection between the distant refiner and its antecedent (A…a), rather than merely marking a contrastive discourse relation (…B-contrast-a).
Nevertheless, we should also consider an additional and/or simultaneous interpretation.It is possible that between refiners a later frontal positivity (LFP) appeared to the cataphoric refiners relative to anaphoric refiners, as for (pro)nouns with no preceding antecedent (Filik et al., 2008;Nieuwland et al., 2019).This would be consistent with our interpretation of frontal positivities appearing to zoom panels in visual narratives which had no antecedents (Cohn & Foulsham, 2020).Under this interpretation, cataphoric panels would be less expected in their position early in the sequence, and thus evoke an LFP, similar to congruous yet unexpected lexical items in language (Federmeier, Wlotko, De Ochoa-Dewald, & Kutas, 2007;Leckey & Federmeier, 2019;Van Petten & Luka, 2012).This interpretation may be supported by the reduced positivity to cataphoric panels appearing to participants with greater comic reading expertise, i.e., fluency would reduce the LFP because such refiners would seem less out of place.
Later epochs also showed evidence of a cost of processing distance dependencies at position 4, manifesting in a leftward negativity in the 500-800 ms epoch to both anaphoric distant refiners and cataphoric distant antecedents compared to panels from sequences with preceding proximal refiners.This negativity was similar between these distance dependencies, despite the physical differences between them.Similarly, comparison between refiners showed that anaphoric distant refiners-the only refiners making a distance dependency-evinced a posterior negativity persisting across the 500-1100 ms epochs.The leftward distribution of this effect at position 4 is also similar to other lateralized negativities shown to processing of narrative structure (Cohn, 2020b;Cohn et al., 2014;Cohn & Kutas, 2017;Cohn, Paczynski, et al., 2012;Silva et al., 2019), and to left anterior negativities shown to distance dependencies (Barkley et al., 2015;Kluender & Kutas, 1993;Kutas et al., 2017;Kwon et al., 2013) and anaphoric relationships (Almor, Nair, Boiteau, & Vendemia, 2017), although these precedents sometimes are earlier in latency, more anterior, and/or more phasic.Under this interpretation, this leftward negativity may index the working memory costs of associating distance connections with prior information (Barkley et al., 2015;Kluender & Kutas, 1993;Kutas et al., 2017;Kwon et al., 2013), a cost that does not differentiate here between the cataphoric versus anaphoric directionality of the referential dependency.
To summarize these results: At position 4, anaphoric distant refiners evoked a reduced N300, and then a sustained anterior negativity.Yet, both distance dependencies-the anaphoric distant refiner and the cataphoric distant antecedent-evoked centroposterior negativities compared to panels in proximal sequences.When comparing across refiners, anaphoric proximal refiners evinced a reduced N300 and N400, yet anaphoric distant refiners showed no such reduction compared to cataphoric refiners.Both anaphoric refiners subsequently evoked a sustained anterior negativity consistent with an Nref to the coreference to earlier antecedents, while anaphoric distant refiners alone sustained a centro-posterior negativity indicating the cost to processing the distance dependency.Such neural responses are consistent with those for both anaphora and distance dependencies in language.

General discussion
This paper explored the constraints placed on visual narratives for panels that zoom-in on information in another panel (refiners) and those that build a spatial inference from disparate entities (e-conjunction).Though refiners and anaphora in language use different representations, we drew an analogy between modalities because refiners involve an indexical referential relationship to an antecedent.On the basis of this analogy, we thus hypothesized that, (1) structural preferences across both order and distance constrain refiners in relation to their antecedents in an analogous way to linguistic anaphora, and (2) that these constraints manifest in (neuro) cognitive processing similar to language.Both hypotheses were supported by our experiments.
Our most persistent finding is a preference for refiners that followed their antecedents (anaphoric) over those that preceded them (cataphoric).In Experiment 1, choices related to order had the strongest preferences across all comparisons, where anaphoric positions were chosen over cataphoric ones, and these were also the fastest choices.In Experiment 2, anaphoric refiners were viewed faster than cataphoric refiners, attributed to their repetition of information in a prior antecedent.In addition, downstream viewing times were greater to those in sequences with cataphoric refiners than those following anaphoric proximal refiners, implying that such comprehension incurred spillover costs.Finally, in Experiment 3, anaphoric proximal refiners evoked an attenuated N400 compared to cataphoric refiners, implying a benefit of the immediate reactivation of zoomed information.In addition, anaphoric refiners of both distances were more negative across sustained later time windows (500-1100 s) compared to cataphoric refiners, implying either that working memory processes were more involved with connecting anaphoric refiners to their antecedents (negativity) or that cataphoric refiners incurred costs as a zoom with no preceding antecedent (positivity).
We also found that refiners that were proximal to their antecedents are preferred over distant ones.In Experiment 1 this manifested as stronger preferences for proximal positions for refiners than distant ones.In Experiment 2, longer viewing times appeared to panels following anaphoric distant refiners than to those following refiners of other sequences, suggesting a spillover cost to the distance dependency of refiner to antecedent.Anaphoric distant refiners also had no attenuation of an N400 compared to anaphoric proximal refiners, implying that they received less benefit of following their antecedent.Anaphoric distant refiners at position 4 of the sequence evoked a greater negativity that sustained from 300 ms past 1100 ms, which also manifested as a greater negativity when compared to other refiners.This effect perhaps explains the spillover of viewing times at the subsequent position 5 in Experiment 2, where the prior refiners were viewed for only ~ 690 ms.In addition, at the fourth panel position, both anaphoric distant refiners and the antecedents of cataphoric distant refiners evoked a leftward centro-parietal negativities, suggestive of a cost to the distance dependencies.
Altogether, these findings imply constraints on refiners in visual narratives.An order constraint is that refiners should follow rather than precede their antecedents (anaphoric > cataphoric), while a position constraint motivates them to remain closer to their antecedent than further away (proximal > distant).While these constraints both exert pressure, the preference for anaphoric order appears to be stronger than that for proximal positions.These constraints and their relative costs are consistent with those found in language processing, thus implying general mechanisms across modalities governing the flow of meaningful sequential information.Phrased in modality-general terms, anaphoric relationships receive a benefit from information flowing from more informative representations to less informative ones (Hofmeister, 2011;Nieuwland et al., 2019), while greater distance between elements incurs greater costs (Gibson, 2000).
Both our behavioral and neurocognitive findings imply such parallels between the visual and verbal modalities.Experiment 2 showed that constraints on distance and order modulate self-paced viewing times similar to findings of reading times of anaphor and distance dependencies in verbal language.Experiment 3 added further detail to these parallels, by showing that similar electrophysiological responses occur to the modulations between distance and order in refiners as have been observed in studies of referential processing in language.These similarities manifest primarily as a reduced negativity (N400) to anaphoric proximal refiners compared to anaphoric distant or cataphoric refiners, and to sustained negativities (SAN/Nref) to refiners in each sequence position, with greater negativities shown to anaphoric than cataphoric refiners.These ERP components imply not only analogous structural constraints, but possibly also similar neurocognitive mechanisms.This work thus adds to prior observations of similar processes evoked by visual narratives and language (Cohn, 2020b;Loschky et al., 2020;Magliano et al., 2015;Robertson, 2000).Further research should examine the degree to which constraints on anaphora (e.g.Almor, 1999) may apply beyond the verbal modality.
Nevertheless, given that these findings imply preference for proximal refiners over distant ones, the question arises why corpus analyses suggest that anaphoric distant refiners appear more prevalently than proximal refiners across most all cultures' comics (Cohn, 2019).Our proposal is that, though it incurs costs, refiner displacement is a constructional narrative pattern (Culicover & Jackendoff, 2005;Goldberg, 1995).This pattern functions to build narrative tension, drawing on a back-and-forth alternation pattern (Bateman & Schmidt, 2012;Cohn, 2019), but does so in a way that belies the semantic preferences for the elements involved (Asher, 2008;Asher & Lascarides, 2003).This echoes the results in Experiment 1, where anaphoric distant refiners obtained greater preference than other less preferred structures, but also required the greatest time for participants to make this decision-reflecting a weighing of their atypical semantics with their status as a narrative construction.Thus, this construction uses a stored narrative sequence, with a non-isomorphic mapping to a parallel semantic structure (Culicover & Jackendoff, 2005;Jackendoff, 2002), and our experimental results support the tension between these component structures.
Altogether, a constructional approach to visual narrative implies that more specificity is required in studying visual narratives than broad appeals to organizing principles claimed to act in a uniform way across (visual) discourse (Bateman & Wildfeuer, 2014a, 2014b;Loschky et al., 2020;McCloud, 1993).Visual narratives are not an anything-goes sequence of pairwise relations between images, nor are they motivated by knowledge of perception and visual events alone (Zacks, 2014).Rather, visual narratives are stitched together from specific constructional schemas guided by constraints, with varying degrees of entrenchment in readers' visual lexicons (Bateman & Schmidt, 2012;Cohn, 2019Cohn, , 2020a)).Thus, like research on syntactic structure, the study of visual narrative sequencing needs to identify and explore these specific constructions.Such specificity can in turn inform us about their organizational principles, and whether they implicate constraints that operate similarly on sequencing across domains.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 3 .
Fig. 3. Example stimuli from Experiment 1.Participants were asked to choose a location to place the zoom panel in the sequence.This example used an "_AB_" choice that would result in either a cataphoric proximal refiner (1, cf.Fig.2b) or an anaphoric distant refiner (2, cf.Fig.2c), both of which would be left-branching.

Fig. 4 .
Fig. 4. Choice frequency and response times for where participants chose to place refiners relative to panels depicting individual A and B characters.Chosen placements created narrative patterns where refiners were anaphoric (A) or cataphoric (C) or proximal (P) or distant (D) from their antecedents, within sequences that created either left-(L) or right-(R) branching structures.Dots show the frequency of choices in each pair, along with the response time indicated by height of the red dot and error bars.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 6 .
Fig. 6.Grand average ERPs for panels at ordinal position 4 across factors of Order (anaphoric, cataphoric) and Distance (proximal, distant) for a) waveforms averaged across regions of interest, and b) topographic voltage maps for the differential effects between sequence types.Highlighted areas show epochs under analysis.
Fig. 7. Grand average ERPs for refiner panels across factors of Order (anaphoric, cataphoric) and Distance (proximal, distant) for a) waveforms averaged across regions of interest, and b) topographic voltage maps for the differential effects between sequence types.Highlighted areas show epochs under analysis.