1.1. Previous research on non-personal pronouns and demonstratives

Understanding the different roles and representations of non-personal and demonstrative pronouns has been notoriously difficult, since their functions and uses overlap considerably. Some researchers claim that “it” and “this”/“that” “are indistinguishable with respect to the description they provide for the intended referent (an inanimate object)” (Ariel, 2001, p.29). Others claim that the interpretation of “it” and “this”/“that” depends largely on discourse status, such that non-personal pronouns are used for topics and/or salient referents, and demonstratives are used for non-topical but activated content (e.g., Gundel et al., 1993; see also Grosz, Weinstein & Joshi, 1995; Grosz & Sidner, 1986; see Bosch, Rozario & Zhao, 2003, for a similar argument for German d-pronouns).

Halliday (1985) proposes a system in which demonstrative pronouns establish reference to a specific token, whereas “it” specifies a non-specific token; this model was extended by Strauss (2002), who proposes a gradience in focus from “this” (high focus, important referent) and “that” (medium focus) to “it” (low focus, unimportant referent). According to these systems, the choice of a particular pronoun depends upon how much attention a speaker is asking the interlocutor to pay to the particular referent. In a similar vein, some researchers have simply claimed that bare demonstratives like “this” or “that” refer to anything that best fits all the cues which a “reasonable and attentive addressee will take the speaker to be exploiting” (Wettstein, 1984: 73; cited in Smit, 2012).

Finally, there is also a claim that “it” is sensitive to syntactic prominence, or grammatical category, while demonstrative pronouns like standalone “that” are more likely to refer to complex/composite entities (Brown-Schmidt, Byron, & Tanenhaus, 2005). In their study, Brown-Schmidt et al. (2005) found that when people were told “Move the cup onto the saucer. Now move it onto the table”, they were more likely to move only the cup; but when they were told to “move the cup onto the saucer. Now move that onto the table”, people were more likely to move the composite object cup+saucer, which has no unified linguistic antecedent.

This finding that comprehenders tend to resolve demonstratives as referring to more complex entities than pronouns is also confirmed by Çokal, Sturt and Ferreira (2016), who found that demonstratives tend to refer to propositions, and simple pronouns to objects linguistically encoded in noun phrases (NPs). In an eye-tracking-while-reading study, they found longer reading times when the non-personal pronoun “it” referred to a proposition than when the demonstrative pronoun “this” referred to a proposition; the pattern was reversed for reference to an object NP.

Recent work employing the sentence continuation method — using both constructed texts and snippets of naturally occurring stories — found that people have a strong tendency to use “this” for event reference, and “it” for object reference, modulated both by verb class and intention to re-mention (Loáiciga, Bevacqua, Rohde & Hardmeier, 2018). The observation that reference to events tends to be accomplished with demonstrative pronouns received further support from recent corpus studies showing that almost three-quarters of demonstratives in dialogues refer to events, while only about 5% of non-personal pronouns do (Evans, 2001; Müller, 2007; Poesio, 2015).

Put together, these studies — in addition to corroborating data from other corpus analyses (e.g. Gundel et al., 1993, 2005) — seem to demonstrate two things: First, simple non-personal pronouns tend to refer to easily accessible referents encoded in prior discourse by noun phrases; and second, there is evidence for form-specific constraints (Kaiser & Trueswell, 2008): Demonstrative pronouns trigger comprehenders to construct an antecedent from previous context, one that does not necessarily have to have a simple NP antecedent. That is, the form itself (simple non-personal pronoun “it” vs. demonstrative pronouns “this” and “that”) provides valuable cues to the comprehender about what kind of referent the speaker intends, even if these cues are probabilistic.

These data fit with a theoretical proposal by Elbourne (2008), who analyzes demonstratives as denoting ‘individual concepts’ packaged as definite descriptions. Crucially, demonstrative pronouns in this model also introduce existence and uniqueness presuppositions, just as strong definite determiners do (Abbott, 2004; Strawson, 1950). That is, “I like this” presupposes that there exists something to like, and that this ‘something’ is uniquely identifiable in the context (just like “I want to eat the raspberry macaron” triggers the presupposition that a particular unique raspberry macaron exists), and can be found in the visual/spatial context (for instance, among the other macarons in the box).³

In this theory of demonstratives, Elbourne (2008) argues that “that” and “this” are forms with a relational component, for instance, distal and proximal factors: It has been argued that “this” tends to refer to things closer in context, and “that” to things further away (e.g. Kruisinga, 1925–32; Quirk, Greenbaum, Leech & Svartvik, 1985, and many others; see Scott, 2013, for a detailed analysis distinguishing the form-based preferences of both demonstrative pronouns.) Demonstratives also take a property as argument, and use both the property and the proximal/distal cue to map onto a specific referent (Reuter & Lew-Williams, 2018). For instance, in the sentence “This is the best macaron”, the comprehender will use the proximal cue and combine it with the property of being a macaron; ideally, this way the comprehender finds the referent that the speaker had in mind (presumably a very delicious macaron nearby.) Elbourne (2008) does not explicitly discuss reference to events, and how exactly the comprehender may map a demonstrative pronoun to a chunk of linguistic or non-linguistic conceptual structure is unclear; but we agree with the analysis that the demonstrative pronoun identifies an individual concept by means of reference (Loar, 1976; see also O’Madagain, 2020, for a compatible approach).

We argue furthermore that it is the function and purpose of the demonstrative to bundle a potentially complex, diffuse set of conceptual structure into such an individual concept, which the linguistic discourse can access and use down the road (cf. Grosz, 2018). Thus, in an extension of this model, we argue in the spirit of Wiese and Maling (2005) that demonstratives can serve as ‘universal bundlers’ for complex concepts, such as events.

1.2. A proposal: demonstratives as ‘universal bundlers’

We propose an approach to demonstratives that views them as potential markers of a conceptual process that bundles a chunk of conceptual structure, and marks that chunk linguistically, such that the bundle can be referred to in the ongoing discourse. In this view, demonstratives tend to accomplish more, and have different goals, than simple pronouns.

Simple pronouns like “he” or “it” are indices that typically link the pronoun to an easily accessible noun phrase in the context, provided that this noun phrase adheres to the constraints posited in the relational and classificatory components of the pronoun (including discourse-level, lexical-level, syntactic, and semantic constraints). This is a complex process, as evidenced by the pronoun resolution literature; yet simple pronouns are still usually properly co-referential with a noun phrase in the discourse, except in the occasional case of an unheralded pronoun (Greene, Gerrig, McKoon & Ratcliff, 1994), or other rare exceptions.

Our account argues that the processes of simple pronoun resolution and demonstrative pronoun resolution, given the same context, are fundamentally different (for precursors of this idea, see e.g., Hankamer & Sag, 1976; Jackendoff, 2002, and probably others): Unlike simple pronouns, demonstratives serve as triggers for bundling chunks of conceptual or linguistic structure into an individual concept, such that this individual concept can serve as a referent for further discourse. This bundling procedure can be purely conceptual in nature, and the demonstrative pronoun is simply the linguistic marker for this ‘universal bundling.’⁴

We propose that non-personal simple pronouns like “it” and demonstrative pronouns like “that” have different form-specific constraints in English (see Kaiser & Trueswell 2008 for the form-specific multiple-constraints framework), and employ different psycholinguistic mechanisms to go from index to interpretation: Whereas “it” has a bias to quickly attach to the first noun phrase that satisfies both its relational and classificatory component, “that” can function as a ‘universal bundler’ that makes a chunk of conceptual structure (whether linguistically encoded or not) accessible to linguistic discourse.⁵ In other words, demonstratives can serve as universal feature bundlers that allow the subsequent linguistic structure to further address the content of these bundles (or, in Lewis & Vasishth’s 2005 terminology, the content of these ‘chunks’).

This idea of a linguistic element serving as linguistic marker and trigger for a conceptual operation is obviously not new. Other ‘universal machines’ have been introduced in linguistics and philosophy to capture mapping functions that take conceptual objects as input, and yield continuous substances as output, or vice versa. The examples in (3) illustrate cases of the ‘universal grinder’, ‘universal sorter’, and ‘universal packer,’ all of which are conceptual operations triggered by mass and count syntax, respectively (Bunt, 1985; Pelletier 1975; Pelletier and Schubert, 1989; examples from Wiese & Maling, 2005):

We propose that demonstratives are linguistic markers, analogous to mass or count syntax, of a conceptual operation that takes a conceptual structure as its input. This operation, in English, can only apply to conceptual structures that satisfy the demonstratives’ form-based relational and classificatory components. Its output is a bundle of conceptual structure (an ‘individual concept’) in a linguistic form, which can serve as a definite description of that individual concept in the ongoing linguistic discourse (see Figure 1).⁶

Figure 1

Illustration of conceptual operations in two ‘universal machines’, after Wiese & Maling (2005).

For instance, an informal, simplified outline of the conceptual structure for the situation depicted on the lower left in Figure 1 could roughly be sketched as Example (4):

Note that this sketch does not take into account the social, visual and spatiotemporal complexities that the static picture in Figure 1 implies: The children’s and woman’s facial expressions and gaze, inferences about the persons’ ages and relations among each other, the architectural style of the kitchen in the picture, inferences about its location and inhabitants, and so on. All of these additional aspects of the event could be, in principle, included in a finer-grained representation of Figure 1. Important to note is that out of this infinitely rich conceptual structure, a speaker can uniquely target any structure and substructure using a bare demonstrative:

The referent of Example (5) is necessarily vague without more context; “that” can bundle anything from the style of the cabinets to the mess on the counter. However, in the absence of prior discourse, once the speaker includes more lexical content to restrict the classificatory component of the pronoun, identification of the referent is easier:

Thus, demonstratives tend to bundle up and refer to eventive or otherwise complex antecedents that are not encoded with a noun phrase, based on these observations and on the evidence from Brown-Schmidt et al. (2005) and Çokal et al. (2016). Hence, we propose the following form-based classificatory and relational properties of English demonstratives:

1. The classificatory component of standalone demonstratives specifies that the search for a referent needs to be restricted to a non-animate entity.

2. The relational component of standalone demonstratives specifies that the referent is not immediately accessible in context.

The classificatory component’s restrictions are intuitively quite straightforward and easily verifiable. For instance, *“I like Nils_i. This_i has such a sunny smile” is ungrammatical in English (even though the use of demonstratives in conjunction with copula verbs may be acceptable for some speakers, i.e., ?“I like Nils_i. This_i is such a happy kid”). We do not discuss this component further.

The relational component’s specification can surface in varying ways. Linguistically, either the conceptual structure to be bundled up is (i) far away in the discourse (Çokal, Sturt, and Ferreira, 2014), or it is (ii) far away from upper layers in discourse structure, i.e., it is neither the current focus nor the current topic in discourse (Webber, 1989), and is less salient on the level of discourse (Gundel et al., 1993) or the level of argument structure (Chafe, 1976; Brennan, Friedman, & Pollard, 1987).

But the referent of a demonstrative does not have to be linguistically expressed, as evidenced by examples (5) or (6) above, drawn from Figure 1. It can also be a percept in the visual or auditory domain, or another non-linguistic and purely conceptual structure, as (7) shows (e.g., Kaplan, 1989; Jackendoff, 2002):

Crucially, demonstratives can not only bundle up non-linguistic conceptual structures, but also conceptual structures that have been encoded linguistically, for instance, when the conceptual structure to be bundled up is not an easily co-referential noun phrase. In the case of Brown-Schmidt et al.’s (2005) study, this bundle contained both the cup and the saucer. The data from their study provides evidence that demonstratives may be used as a cue to gather and bundle up concepts expressed by noun phrases. But how do people refer back to things that are realized linguistically but not in the form of noun phrases, such as in (6a) and (6c)? To answer this question, we turn to event descriptions, such as in “The friends visited Paris”, “The hikers explored the forest”, or “Adam heated the lasagna”.

1.3. Current studies and predictions

In this paper, we use reference to objects (linguistically realized as NPs) and events (conveyed by an entire clause) to investigate whether non-personal pronouns like “it” and stand-alone demonstrative pronouns like “this” or “that” access different cognitive mechanisms in reference resolution in English.

We use sentence pairs like in (8), where a context sentence sets up a potential referent for the pronoun that is present in the next (critical) sentence: “it” refers to “lasagna”; “that”, to the act of making the lasagna. We use “it” and “that” because they are distributionally more similar to each other than “it” and “this” (Strauss, 2002).

Specifically, we propose that non-personal pronouns are interpreted as coreferential with (salient) lexical item that satisfies the pronoun’s form-specific relational and classificatory constraints: inanimate, grammatically singular noun phrases (“lasagna” in (8)). We propose that, in the context of our experiment, this is a process in which only the linguistic surface needs to be implicated (cf. Hankamer & Sag, 1976): In order to determine the referent of “it”, it is sufficient to access form-based information of the candidate words in context. In languages with grammatical gender, this information may be morphological information (Cacciari, Carreiras & Barbolini Cionini, 1997); but also, lexemic properties such as length and frequency of a word (Simner & Smyth, 1999; Duffy & Rayner, 1990). Thus, we predict that “it” should be sensitive to frequency and length of the preceding linguistic material resolution (although see Egusquiza, Navarrete & Zawiszewski, 2016, and Lago, 2014, for failure to find frequency effects with personal pronouns).

We also predict that the demonstrative “that” should be sensitive to the same surface-form-based features, because these features need to be accessed for any kind of reference. Overall, we predict that after encountering “it” or “that” at the start of the critical sentence, reading times – which reflect processing ease – will reveal effects of the antecedents’ surface properties.

We further predict that “that” will lead to slower reading times than “it.” This is because under our approach, “that” crucially differs from “it” in that only “that” is accessing and bundling up complex conceptual structures – processes which can be assumed to carry a cognitive cost. Thus, we predict that “that” should be read slower than “it” throughout, above and beyond the difference in orthographic length of the pronouns.

Additionally, we predict that “that” should be uniquely sensitive to higher-level conceptual features, such as the complexity of a concept. Thus, we manipulate the conceptual complexity of an event (as discussed below) that a subsequent demonstrative “that” will refer back to.⁷ In particular, we predict that (i) more complex events will lead to faster reading times than less complex events for sentences where “that” is used to refer to the event, whereas (ii) we do not predict effects of event complexity for sentences where “it” is used to refer to the event. This prediction originates from literature suggesting that semantically rich representations lead to faster re-access than semantically poor representations (Fisher & Craik, 1980; Craik & Tulving, 1975; Gallo, Meadow, Johnson & Foster, 2008; van Gompel & Majid, 2004; Heine et al., 2006a, b; Hofmeister, 2011; Karimi & Ferreira, 2016).

Finally, we also make an additional prediction that when non-personal pronouns are subsequently specified by event-denoting adjectives (such as “It was very adventurous” or “It was quite laborious”), comprehension should slow down at the adjective, due to an effect of mismatched expectations; likewise, we expect the same when demonstratives are subsequently specified by object-denoting adjectives (such as “That was very small” or “That was quite pretty”). We refer to the difference between adjectives like “adventurous” (event-denoting) and “small” (object-denoting) as adjective bias. Observing these kinds of slow-down patterns would indicate that readers consider an event reference more when they have read a demonstrative, and an object reference more when they have read a non-personal pronoun. Violations of these expectations would result in a type mismatch, as an interaction of adjective bias with pronoun type. However, we issue this prediction with caution for two reasons: In order to limit the length of the experiment, we introduced such a mismatch only for half of our experimental items, and thus, power is significantly reduced for this analysis. Second, the adjectives were in sentence-final position, which has been associated with complex sentence-wrap up effects (e.g., Warren, White & Reichle, 2009).

In what follows, we present two experiments, each replicated once with a slight difference in stimuli, to test these hypotheses. Both experiments allow us to test the prediction that “that” is read more slowly than “it”, and that adjective bias may interact with pronoun type. Experiments 1a and 1b investigate whether both non-personal and demonstrative pronouns are sensitive to surface features of the linguistic context, including those of the potential referents. Experiments 2a and 2b test the prediction that only demonstrative pronouns are sensitive to higher-level features of the linguistic context.

2.2. Results

We excluded 34 participants because of low comprehension-question accuracy (<75%), or because their median reading times were below 200 ms or above 1500 ms. The accuracy of comprehension questions in the remaining 166 participants was 92% (range = 79% ~ 100%). Reading times above 2000 ms (<.5% of the data) or below 100 ms (<1% of the data) were also excluded.

Figure 2 shows log reading times for each region in Experiment 1, and Table 1 shows the mean reading times with Standard Errors. A summary of the regression results is shown in Table 2. While our predictions only pertain to the critical sentence, we report reading times in each region (including the context sentence) for completeness’s sake.

Figure 2

Region-by-region reading times in the critical sentence, on log scale, in Experiment 1a.

In the context sentence, there was a spurious effect of pronoun in the very first region, the subject NP (“the hunters”). Spurious effects are very common in self-paced reading experiments (e.g. Omaki, Lau, Davidson White, Dakan, Apple & Phillips, 2015; Meng & Bader, 2020; among many others). We classify effects as spurious when they fulfil two criteria: When they could not have been introduced by our manipulation, i.e. because they occurred before the manipulation, and when they do not consistently occur between experiments or across subsequent regions.

As predicted, starting in the object noun region (“the forest/jungle”), people read faster starting at a high-frequency noun (“forest”), than after a low frequency noun (“jungle”). This main effect of frequency remains significant over three contiguous regions: the object noun region (the region where frequency was directly manipulated) in the context sentence, and the pronoun region and the copular verb region in the critical sentence. It is marginal on the subsequent adverb region. Furthermore, we also find a main effect of pronoun type, with “it” conditions being read faster than “that” conditions, at the copular verb that immediately follows the pronoun as well as at the immediately following adverb. There are no interactions involving frequency and pronoun type anywhere in either of the two sentences.

We also analyzed the adjective region (“adventurous”), with adjective bias as an additional fixed predictor: Here, no main effects were significant. We also did not find any significant interactions of adjective bias with complexity or pronoun type (all ps > .15).

3.2. Results of Experiment 1b

Of 200 initial participants, we excluded 29 based on the same criteria as in Experiment 1a: Either comprehension question accuracy of less than 75% (N = 11), median reading time (across all regions) of less than 200 ms (N = 19) and more than 1500 ms (N = 0). As before, we also excluded trials with reading times above 2000 ms (2.5% of observations) or less than 100 ms (<1% of observations). The accuracy of comprehension questions in the remaining participants was 94% (range = 77% ~ 100%).

Figure 3 shows log reading times for each region in Experiment 1b; Table 3 shows the mean reading times with Standard Errors, and Table 4 shows regression results on raw reading times over the critical regions.

Figure 3

Region-by-region reading times in Sentence 2, on log scale, in Experiment 1b.

We find a spurious effect at the Context Sentence’s subject as a main effect of frequency. Predicted effects of our frequency manipulation started at the object noun: High frequency nouns were read faster than low-frequency nouns. This effect of frequency continued and did not subside until the end of the second sentence; specifically, pertaining to our predictions, starting in the pronoun region (“that/it”), people read faster after a high-frequency noun (“forest”), compared after a low frequency noun (“jungle”), with a significant main effect of frequency. Importantly, at the following copular (“was”), at the adverb (“very”), and, marginally, in the sentence-final adjective, we also found a main effect of pronoun type, with people reading faster after “it” than “that.”

We also found a marginal interaction of pronoun type with frequency, at the pronoun region, with an unpredicted slightly larger effect of frequency for demonstratives than for pronouns. Again, we did not find any significant interactions of adjective bias with complexity or pronoun type (all ps > .34).¹¹

5.2. Results

We excluded 37 participants because of low question accuracy, or because their median reading times were below 200 ms or above 1500 ms. Reading times slower than 2000 ms (<2% of observations) or faster than 100 ms (<2% of observations) were excluded as well. The accuracy of comprehension questions in the remaining participants was 94% (range = 75% ~ 100%).

Figure 5 shows log reading times for each region in Experiment 2a; Table 5 shows the mean reading times with Standard Errors, and Table 6 shows model results for the critical regions. Before the pronoun region, no significant effects were found (ps > .05), except for a main effect of Complexity at the object NP following the manipulated verb, but this effect had subsided by the next region. As predicted, and consistent with results from Experiments 1a and 1b, we found a main effect of pronoun type starting in the pronoun region (“that/it”), which remained significant throughout the rest of the trial: “it” conditions were read faster than “that” conditions. No other effects were significant, except for a main effect of complexity at the sentence-final adjective; we did not find any significant interactions of adjective bias with complexity or pronoun type (all ps > .38).

Figure 5

Region-by-region reading times in the Critical Sentence, on log scale, in Experiment 2a.

6.2. Results

From the initial set of 200 speakers, we excluded 25, because of low accuracy on comprehension questions (N = 3), or due to median reading times being too fast (N = 20) or too slow (N = 2). The accuracy of comprehension questions in the remaining participants was 94% (range = 78% ~ 100%). Again, we also excluded individual trials based on reading times: Longer than 2000 ms (<3% of observations) or shorter than 100 ms (<2% of observations).

Figure 6 shows log reading times for each region in Experiment 2b; Table 7 shows the mean reading times with Standard Errors, and Table 8 shows results of the regression.¹⁵ We find a spurious main effect, and spurious effects of interactions with pronoun type in four regions of the context sentence. However, as expected, and as found in Experiment 2a, we also found a main effect of complexity at the manipulated verb (“entered/explored”) and the object NP. Reassuringly, this effect had subsided in the spillover region, and only reached marginal significance at the pronoun at the start of the critical sentence.

Figure 6

Region-by-region reading times in the Critical Sentence, on log scale, in Experiment 2b.

In the auxiliary region (“was”) and the adverbial region (“very”) of the critical sentence, we find a main effect of pronoun type, but no effect of complexity, and no interaction, in line with our other experiments. This time, we also find the predicted interaction of adjective bias with pronoun type at the final adjective (marginal, β = .03, p < .07), a main effect of complexity (β = .03, p < .05), and a three-way interaction between pronoun type, complexity, and adjective bias (β = .07, p < .02).

8.1 Sensitivity to surface properties

In the first pair of self-paced reading studies (Experiment 1a and 1b), the data show that both non-personal and demonstrative pronouns were read faster after nouns that were shorter in length and higher in frequency than after nouns that were longer and less frequent. Importantly, these results also surface when there is a delay between the referent and the pronoun, such that the reduction in reading times cannot merely be taken as a spillover effect from the nouns themselves, but rather can be attributed to the reference resolution process itself, supporting our model of demonstratives as bundlers of conceptual structure as opposed to simple linguistic anaphora devices. In addition to this, these findings provided the baseline for Experiments 2a and 2b, which asked whether “that”, and not “it”, is uniquely sensitive to higher-level conceptual features of its referent.

8.2 Sensitivity to conceptual properties

The second part of our hypothesis that the bare demonstrative “that” tends to bundle up complex conceptual structures, whereas “it” does not necessarily do so, results in two predictions: First, “that” should result in longer reading times throughout compared to “it”, and “that” should be uniquely sensitive to referential complexity.

The first prediction was confirmed in all four experiments: The demonstrative “that” resulted in longer reading times than “it”, and this main effect was robust, stable, and replicable. This pattern is very likely not due to orthographic differences, since we also analyzed the data residualized over the length of the pronouns, and still found this effect. On the contrary: Since both forms (non-personal pronouns and demonstratives) are on the very high end of the lexical frequency spectrum, finding any effect on the pronoun region is striking, given that due to the word frequency and the word length effect, reaction times to short, high-frequency words tend to stick to floor level (see e.g., Morton, 1970; Kuperman, Drieghe, Keuleers & Brysbaert, 2013; Dirix, Brysbaert & Duyck, 2019, and others for data from many behavioral paradigms). Thus, these results are a powerful demonstration of the different reading behaviors induced by non-personal pronouns and demonstratives.

To test the second prediction, we conducted Experiments 2a and 2b, which manipulated potential referents’ conceptual complexity. Specifically, we manipulated event complexity: We compared sentences with more complex events to sentences with less complex events. We expected the demonstrative “that” to pattern differently from “it” as follows: Based on prior research, and our own previous data, we expected an interaction such “that” would be read faster in conditions with more complex events than in conditions with simpler events. This is because “that” tends to refer to the event, and complex events have been shown to be more easily retrieved the more complex they are (Hofmeister, 2011; see below).¹⁶ While this second prediction was not borne out, the first was, showing that demonstratives are processed differently from personal pronouns, affecting reading times overall.

8.3 Our results in light of previous literature

Let us briefly compare our results to other experiments on antecedent frequency effects on pronoun processing, because our results may, at first blush, seem surprising given prior work: Several recent studies have reported that less frequent antecedents lead to faster reading times at the pronoun (van Gompel & Majid, 2004), or to no reliable differences at the pronoun (Egusquiza et al., 2016; Lago, 2014). However, each of these studies is crucially different from ours in several aspects. First and foremost, van Gompel & Majid (2004) as well as Lago (2014) used animate antecedents (“the arsonist/the criminal”), and likely more importantly, unambiguous possessive determiners (“his bag”) to investigate effects of frequency. Second, they measured effects in eyetracking-while-reading paradigms. And third, they used frequency as a proxy for saliency, on the assumption that all low-frequency items are more salient than high-frequency items.

In contrast, we used a standard masked self-paced reading paradigm, and measured reading times at a bare (non-human-referring) pronoun which was, in principle, compatible with at least two different referent types: the sentential object alone, such as “forest” or “jungle”, or the whole event (“The hikers entered/explored the forest/the jungle.”) We also used items that were close synonyms, such as “forest” and “jungle”, and not as conceptually far apart as many of van Gompel & Majid’s (2004) as well as Lago’s (2014) stimuli (pairs in these papers were, for instance, “student” vs. “vagrant”, or “doctor” vs. “envoy”, which may have entailed not only a differential in frequency, but also in register, pragmatics, and semantic associations and features.) In light of these differences, direct comparisons between experiments are presumably not meaningful.¹⁷

In contrast, Hofmeister (2011) explicitly manipulated conceptual complexity. He found that people were faster reading “banned” in sentences like (17b) than in sentences like (17a); that is, the more complex NP “alleged Venezuelan communist” in (17b) was easier to retrieve from working memory and integrate in the argument structure than the less complex NP ‘communist’ in (17a):

In Hofmeister’s (2011) stimuli, ‘alleged Venezuelan communist’ is indeed semantically richer than ‘communist’, but it is also a longer, more complex noun phrase (and retrieval effects could potentially be due to longer encoding time, Karimi, Diaz & Wittenberg, 2020); whereas in our study, the only variation between the verbs was one of explicitly controlled conceptual complexity. Hofmeister’s Experiment 2, which replaced semantically specific referents like “soldier” with non-specific referents like “person” and is in some sense similar to ours, found only marginally significant retrieval effects.

Even without an effect of conceptual complexity, however, our data can be taken as evidence that “that” and “it” result in significantly different processing. We propose a model of how a bundling process may be triggered by demonstratives like “this” or “that”, and we view this proposal as an extension and unification of approaches to demonstratives that have described their functions in discourse from the perspective of information structure (Ariel 2001, Gundel et al., 1993; Strauss, 2002), or from the perspective of anaphora (Çokal, Sturt, & Ferreira, 2016).

We must stress that the form-specific preferences and mechanisms that we propose here are not deterministic: In English at least, both “it” and “that” can refer to both objects and events. Furthermore, both “it” and “that” can be temporarily ambiguous: “it” could be an expletive (“It is raining”), and “that” could be a demonstrative determiner (“that macaron”; although see Strauss, 2002, for corpus data on how “that” is more than twice as often used on its own as a bare demonstrative pronoun, as opposed to as a modifier to an NP). It seems unlikely that these temporary ambiguities could explain away the results we observed, but we do think that our studies should be extended, possibly into other languages.

Languages other than English also often show a contrast between simple pronouns and demonstrative pronouns, as extensive fieldwork has shown (e.g. Diessel, 1999; Dixon, 2003; Givón, 1978; Himmelmann, 1996); and it must be assumed that they slice the conceptual pie between object and event reference in interestingly different ways than English does (for instance, see Bosch et al., 2003; Grosz, 2018; Kaiser, 2011; and many others for German d-pronouns). To take an example that has caught the attention of psycholinguistic research, Kaiser & Trueswell (2008) investigated the processing of simple and demonstrative personal pronouns in Finnish. Their data indicate that these types of pronouns exhibit different form-specific constraints: Whereas the personal pronouns are sensitive to grammatical role, the demonstratives were sensitive to both information structure and grammatical role. Little is known so far about event or object reference in Finnish, however. It will be interesting to see whether all languages have elements that can function as bundlers, and if so, whether demonstratives (or their equivalent) cross-linguistically show this tendency to be able to refer to conceptual structures wider than simple pronouns.¹⁸

Other open questions concern the limits of bundling. If it is true that bundling is a conceptual process, it should operate under the same working memory constraints as operations on other cognitive units (Baddeley, 2012). In recent years, there have been many promising attempts at integrating research on working memory and language processing. Lewis & Vasishth (2005), a prominent example, proposed a content-addressable memory architecture that is integrated within linguistic theory. Our model of the role of the bare demonstrative “that” as a universal bundler fits squarely within this theory, while also accounting for the non-linguistic, conceptual content that demonstratives bundle up for use by linguistic means.

In sum, in this paper we presented data from two English self-paced reading studies, each replicated once, showing that demonstratives are processed differently from personal pronouns, and that this affects reading patterns throughout the whole sentence. These data can be seen as initial evidence supporting a new, unified model of reference by demonstratives as a process of conceptual bundling, with demonstratives as operators on the interface of language and broader cognition.