Abstract
We examined whether the position of modifiers in English influences how words are encoded and subsequently retrieved from memory. Compared to pre-modifiers, post-modifiers might confer more perceptual significance to the associated head nouns, are more consistent with the “given-before-new” information structure, and might also be easier to integrate because the head noun is available before the modifications are encountered. In four experiments, we investigated whether pre-modified (the cruel and merciless king), and post-modified (the king who was cruel and merciless) noun phrases (henceforth NPs) could induce variations in ease of subsequent retrieval. In Experiments 1, 2 and 3, participants used more pronouns (he), as opposed to full descriptions (the king) to refer to post-modified NPs than to unmodified competitors, but pronominal reference to pre-modified NPs and unmodified competitors did not differ, suggesting that post-modified NPs are more accessible in memory. When the data from all three experiments were combined, we also observed significantly more pronominal reference to post- than to pre-modified NPs, as well as a greater increase in pronominal reference rates between post-modified NPs and unmodified competitors than between pre-modified NPs and unmodified competitors. Experiment 4 revealed faster reading times for words following pronouns that referred to pre- and post-modified NPs compared to unmodified NPs, and also to post- relative to pre-modified NPs. Taken together, our results show enhanced retrieval facilitation for post-modified NPs compared to pre-modified NPs. These results are the first to demonstrate that the linear position of modifications results in measurable processing cost at a subsequent point. The results have important implications for memory-based theories of language processing, and also for theories assigning a central role for discourse status and information structure during sentence processing.
Keywords: Semantic richness, Modifier position, Encoding, Retrieval, Referential processing
Introduction
Language processing involves encoding words, maintaining them in memory, and retrieving them at subsequent points when necessary. For instance, when processing The man drank the juice because he was thirsty, memory representations for the man, juice and other words are created and maintained in memory, and when a retrieval trigger such as he is encountered, the associated representation (i.e., the man) is retrieved from memory to support successful formation of the referential dependency between the pronoun (he) and its “referent” (i.e., the man, Cook, 2014; Dell, McKoon, & Ratcliff, 1983; Gernsbacher, 1989; Gerrig & McKoon, 1998; Lucas, Tanenhaus, & Carlson, 1990; MacDonald & MacWhinney, 1990; Sanford & Garrod, 1989, 2005). A property of most pronouns is that they do not contain much semantic content and therefore the retrieval of the associated representation helps create a coherent discourse representation via the anaphoric link. It is important to clarify that while we talk about pronouns as referring to preceding associated NPs for the sake of simplicity, the reader should bear in mind that an NP itself is a referring expression in the sense that it necessarily refers to a real-world entity or an entity in the interlocutor’s mental model. In this sense, pronouns do not actually refer to the surface NP they are associated with, but rather to the entity the NP refers to.
Previous research has demonstrated that semantically enriching an NP by adding modifying information to it facilitates the subsequent retrieval of that NP from memory. Specifically, prior research has shown that a word is processed more easily if it triggers the retrieval of a modified NP compared to when it requires the retrieval of an unmodified NP. Interestingly, English allows NPs to be both pre- and post-modified and previous studies have employed both modification types to investigate the effect of semantic richness on subsequent retrieval. For instance, using pre-modifiers, Hofmeister (2011) showed that semantically richer NPs such as an alleged Venezuelan communist result in faster reading times relative to unmodified NPs such as a communist, when a subsequent verb (e.g., banned in … who the members of the club banned) triggers the retrieval of the memory representation of the target noun (i.e., communist, also see Hofmeister & Vasishth, 2014). Similarly, using post-modifiers, Karimi and Ferreira (2016a) demonstrated that ambiguous pronouns tend to be interpreted as referring to semantically richer NPs such as the actor who was frustrated and upset compared to unmodified NPs (i.e., the actor). These results have been interpreted as reflecting higher memory activation (i.e., higher accessibility1) for the memory representation associated with richer NPs (e.g., Ariel, 1990; Fletcher, 1984; Gernsbacher & Hargreaves, 1988; Givón, 1983; Gundel, Hedberg, & Zacharski, 1993; Lewis & Vasishth, 2005; Lewis, Vasishth, & Van Dyke, 2006). In other words, semantically richer words seem to result in more robust encoding, which then leads to easier retrieval later in the discourse. Interestingly, the effect of modification is not restricted to language comprehension: Karimi, Fukumura, Ferreira and Pickering (2014) showed that compared to unmodified NPs, post-modified NPs are more likely to be subsequently referred to with pronouns (he) than with repeated nouns (the actor), which is in line with previous research showing that pronouns (and attenuated referring expressions, in general) reflect higher accessibility for the associated memory representation (Arnold, 2001; Brennan, 1995; Fukumura & Van Gompel, 2010, 2011; Stevenson, Crawley, & Kleinman, 1994). Thus, semantic richness has been shown to lead to easier subsequent retrieval during both language production and language comprehension.
An important question that remains unanswered regarding the semantic richness effect is whether the position of modifiers affects how modified NPs are encoded and later retrieved from memory. Languages sometimes provide their speakers with choices as to how to encode a message. For example, a message could be delivered through an active syntactic form (The man drank the juice) or a passive form (The juice was drunk by the man); a previously mentioned noun phrase such as the man could subsequently be referred to with a pronoun (he) or with a repeated noun (the man); a complementizer such as that could be mentioned (I think that the man is thirsty), or be left out from a sentence (I think the man is thirsty); and so on and so forth. Importantly, most, if not all, of these alternative forms result in measurable variations in processing cost, and numerous psycholinguistic studies have investigated the processing costs of these alternative forms. For instance, passive sentences have been shown to be more difficult to process than active sentences (e.g., Bock & Warren, 1985; Ferreira, 2003; see Chomsky, 1986; 1993; 1995); repeated referring expressions result in more processing difficulty than a pronoun when they refer to highly activated NPs (e.g., Almor, 1999; Gordon, Grosz, & Gilliom, 1993), and not realizing the optional complementizer that for less predictable predicates has been shown to lead to processing difficulty (e.g., Ferreira, & Henderson, 1990; Trueswell, Tanenhaus, & Kello, 1993; also see Jaeger, 2010). To the best of our knowledge, the potential processing consequences of the choice between pre-modified and post-modified NPs (e.g., the red car vs. the car that’s red) has not been investigated. As mentioned above, English allows both types of modifications and thus provides an opportunity to investigate this question within the same language.
Motivation and predictions
We expected modifier position to affect subsequent retrieval based on previous research showing that syntactic information (such as subjecthood, c-command, etc.) can restrict the search space for retrieval candidates, thereby reducing interference (e.g., Arnett & Wagers, 2017; Dillon, 2011; Dillon, Chow, Wagers, Guo, Liu, & Phillips, 2014; Kush, 2013; Parker, Shvartsman, & Van Dyke, 2017; Van Dyke & McElree, 2011). Moreover, previous research has demonstrated that the production of pre- and post-modified NPs may depend on how efficiently the modifying information can constrain the potential referent on a visual display (Rubio-Fernández, 2016). If syntactic constraints can influence retrieval, and if the choice between pre- and post-modifying information can influence communication efficiency, it is conceivable that the syntactic choice between pre- and post-modifiers might also influence how NPs are encoded and subsequently retrieved.
There are three possible outcomes: 1. Pre-modified NPs are more robustly encoded in memory than post-modified NPs, 2. Post-modified NPs are more robustly encoded than pre-modified NPs, and 3. Pre- and post-modified NPs do not differ in encoding efficiency. Interestingly, all three possibilities receive support from the current psycholinguistic literature.
Pre-modified NPs might be encoded more efficiently than post-modified NPs because the head of a pre-modified NP is relatively less susceptible to time-based decay. Memory representations are assumed to decay over time, and their susceptibility to decay increases with the temporal distance between when they are encoded, and when they are retrieved (Baddeley, 2000; Chomsky, 1965; Gibson, 1998; Lewis & Vasishth, 2005; Lewis et al., 2006; Van Dyke & Lewis, 2003). Since post-modifiers, but not pre-modifiers, necessarily add more temporal distance between the head noun and the subsequent retrieval site, the memory representation of the head noun is more likely to decay (i.e., lose activation strength) when it is post-modified than when it is pre-modified.
Another reason why pre-modified NPs could be encoded more strongly in memory than post-modified NPs is because pre-modifiers might narrow down the number of possible upcoming head nouns (i.e., reduce the entropy or uncertainty about the head noun), thereby rendering the head noun more predictable (Hale, 2001, 2006; Levy, 2008; note that even if the lexical identity of the upcoming head noun is not predictable, the head of a pre-modified NP is still relatively more predictable than the head of a post-modified NP). This predictability might lead to pre-activation of the features of the upcoming head noun and also to an active maintenance of those features in memory, which might in turn facilitate encoding (of the head noun itself as well as the integration of modifying information) when the head noun is eventually encountered. Since the head noun is revealed before the modifying information in the case of post-modified NPs, such pre-activation and/or active maintenance is unavailable for post-modified NPs, potentially rendering them more vulnerable to decay. In fact, predictability of upcoming information plays a central role in one of the main explanations proposed to account for the observation that semantically rich NPs facilitate subsequent retrieval. Under this account, known as head- reactivation, predictable information is pre-activated and then becomes reactivated as the modifying words are being encoded into memory (Hofmeister, 2011; Lewis & Vasishth, 2005; Lewis, Vasishth & Van Dyke, 2006). This reactivation has been argued to facilitate the incorporation of the extra semantic content with the head noun, and also to prevent the decay of the head noun’s memory representation due to passage of time (Lewis & Vasishth, 2005; Lewis, Vasishth & Van Dyke, 2006). For example, based on the head-reactivation account, pre-modifiers result in higher activation for the head noun because the syntactic category of the head (i.e., “noun”) is pre-activated when the modifying words are encountered and becomes re-activated with each pre-modifying word2, giving the memory representation of the syntactic category multiple activation boosts, and leading to higher ultimate activation of the head noun compared to unmodified NPs. For instance, when encoding a pre-modified NP such as the frustrated and visibly upset actor, the syntactic category “noun” is pre-activated as soon as the frustrated is encountered, because a “noun” syntactic category is highly predictable given the frustrated (note that other features such as “+human” etc. could also be predicted and pre-activated). Importantly, the memory representation associated with the syntactic “noun” category becomes reactivated with each modifying word, preventing it from decay and facilitating the integration of the modifying words with the head noun. As a result, when the head noun is finally revealed, it enjoys a higher level of activation than it would have if the noun were not pre-modified. Also, note that since pre-activation of the head facilitates the incorporation of dependent semantic content (i.e., the modifying information), this mechanism may boost semantic richness as well.
However, some other findings predict a more robust encoding, and therefore an easier subsequent retrieval, for post- than pre-modified NPs. For instance, Dillon, Clifton, Sloggett & Frazier (2017) demonstrated that “at-issue” linguistic content causes more interference during the retrieval of a target item than “not-at-issue” content, suggesting that at-issue information is encoded more robustly in memory. Specifically, Dillon et al. showed that restrictive relative clauses embedded in a main clause (e.g., The butcher asked who the lady who bought Italian ham was cooking dinner for) produce more interference during the subsequent retrieval of the displaced target NP (i.e., “butcher”) than do nonrestrictive clauses containing the same information (The butcher asked who the lady, who bought Italian ham, was cooking dinner for). Dillon et al.’s results are particularly interesting because they show that comprehenders’ perception of the same information might differ based on the discourse status of that information, and that such perceptual variation can influence memory encoding (also see Dillon, Clifton, & Frazier, 2014).
Based on these results, we hypothesize that there might be greater perceptual significance associated with post- than with pre-modified NPs. We argue that this is because, in English, the contents of post-modifying relative clauses tend to be generally more complex than those of pre-modifying adjectives. For example, the more information contained in a modification, the more likely it is to be conveyed with a post-modifying relative clause than with pre-modifying adjectives (“The very thirsty, terribly hungry and deeply agitated man” vs. “The man who was very thirsty, terribly hungry and deeply agitated”). Similarly, post-modifying relative clauses are more variable in content: They might include active clauses (“The man who is wearing glasses”), passive clauses (“The man who was shot”), clauses containing another NP (“The man who kissed the woman”), etc. Moreover, post-modifying relative clauses are more likely to carry more specific information. For instance, although “The thirsty man” and “The man who was thirsty” form acceptable NPs, “The man who was thirsty since morning” forms a grammatical noun phrase, but “The thirsty since morning man” does not. As such, language users might have implicitly learned (through repeated exposure) that post-modifying information is more strongly associated with more complex information, and might therefore allocate more attentional and/or memory resources to them during encoding. As such, just like restrictive clauses, post-modifying relative clauses might be perceived as conveying relatively more important (i.e., “at-issue”) information than pre-modifying adjectives, even though they might not actually carry information that is more important to the current discourse. Nonetheless, this perceived importance might lead to more robust encoding and therefore easier subsequent retrieval of post- than pre-modified NPs. Yet another reason why post-modified NPs might enjoy a greater perceptual salience in memory might be that they already contain a relative pronoun (i.e., who) which might boost the perceptual topicality of the head noun (see Hemforth, Konieczny, & Scheepers, 2000).
A second source of support for the expectation of more efficient encoding of post-modifiers relative to pre-modifiers comes from the information structure of the two modification types. Specifically, some theories and findings suggest that language processing is easier when “given” (i.e., presupposed) information precedes “new” (i.e., focused) information (e.g., Ariel, 1990, 1991, 1996; Benatar & Clifton, 2014; Chafe, 1976; Clark & Clark, 1978; Halliday, 1967; Haviland & Clark, 1974). In the case of modified NPs, the head noun may constitute “given” information and the modifying information may constitute “new” information. This is because the modifier adds new information to the default lexical semantics of the head noun. In other words, the head noun constitutes the core concept to which supplementary (new) information is added. Crucially, post-modified NPs conform to the “given-before-new” principle because the head noun (i.e., the given information) precedes new information (i.e., the modification), whereas pre-modified NPs are not consistent with the given-before-new information structure because the head noun follows the modification. As such, post-modified NPs might be easier to process, integrate, and encode in memory, which in turn may facilitate their retrieval. It is important to mention that from a pragmatics perspective, a modified NP may actually be interpreted as conveying new information regardless of modifier position, and an unmodified NP could be taken as conveying given information. This is because the reason additional information is needed in the first place could be that the referent of a modified NP is not unique enough or sufficiently topical in discourse, hence requiring extra information for clarification. However, note that our critical focus here is between pre- and post-modified NPs and not between pre- or post-modified vs. unmodified NPs. Both pre- and post-modified NPs carry extra information, rendering their discourse status the same. Crucially, however, the information contained in post-modified NPs proceeds from the core head noun (which is supposed to be given information) towards supplementary material (i.e., new information), whereas the information structure for pre-modifiers follows the reverse order (i.e., from supplementary to core information).
A third reason why we might expect a more robust encoding of post- vs. pre-modifying information pertains to the different memory demands associated with processing these two types of modifications. In the foregoing, we argued that active maintenance of the pre-activated features of the head noun might give an encoding advantage to pre-modified NPs. However, it is also possible that such maintenance might tax memory resources and actually complicate encoding. Specifically, and as mentioned above, pre-modifiers need to be maintained in memory until the head noun is encountered, which delays integration. However, no such maintenance is necessary for post-modifiers because the modifying information follows the head noun, and integration can take place immediately. As such, pre-modified NPs might incur a “maintenance cost”, which has been shown to tax memory resources (Wagers & Phillips, 2014)3.
Finally, no difference in the encoding of pre- and post-modifiers is predicted by the distinctiveness account for the effect of semantic richness on subsequent retrieval. Based on this account, semantically richer NPs result in the creation of memory representations that are more distinct compared to unmodified NPs. As such, there is less interference from other items in memory during the retrieval of semantically richer NPs (Gallo, Meadow, Johnson, & Foster, 2008; Jacoby, & Craik, 1979; Hofmeister & Vasishth, 2014; Nairne, 2006). Importantly, if the modifying information is kept constant, both pre- and post-modified NPs should be equally distinct in memory, resulting in no variation in encoding efficiently and therefore later retrieval.
Thus, to contribute to the current literature on how semantically rich concepts are encoded and retrieved from memory, we investigated whether there are processing costs associated with the choice between pre-modifying and post-modifying a noun within an NP. Specifically, we investigated whether the subsequent retrieval ease of the memory representations associated with pre- and post-modified NPs differ from each other. Moreover, given the variety of predictions in the current literature, the results of this study also advances our understanding of the role of perceptual, information structure and memory factors during the encoding of words.
The present study
To answer our questions, we capitalized on the flexibility of English with regard to the position of modifiers. As mentioned above, English allows modifiers to either precede a noun (the cruel and merciless king), or to follow it (the king who was cruel and merciless). Thus, if the retrieval ease of pre- and post-modifiers differ, we can observe these effects within the same language. In this study, we conducted four experiments to investigate whether pre- and post-modified NPs differ in terms of how easily they can be retrieved from memory. In Experiments 1, 2 and 3, we measured the potential effect of modifier position on the retrieval ease of associated NPs during language production. In Experiments 4, we investigated this same question during language comprehension.
Experiment 1
In this experiment, we took advantage of the well-established finding that relatively more accessible (i.e., easily retrievable) NPs are subsequently referred to with more attenuated referring expressions such as pronouns (e.g., he), whereas less accessible NPs are usually subsequently referred to with less attenuated referring expressions such as repeated nouns (e.g., the king). For example, syntactic subjects, animate NPs, and modified (i.e., semantically richer) NPs have been shown to be more likely to be subsequently pronominalized compared to syntactic objects, inanimate NPs, and unmodified (i.e., semantically less rich) NPs, respectively (Syntactic subject vs. object: Arnold, 2001; Brennan, 1995; Fletcher, 1984; Fukumura & Van Gompel, 2010, 2011; Stevenson, Crawley, & Kleinman, 1994); Animate vs. inanimate: Fukumura & Van Gompel, 2011; modified vs. unmodified: Karimi et al., 2014). Since syntactic subjecthood, animacy and semantic richness have all been shown to boost the accessibility of associated NPs (Subjecthood: Brennan, 1995; Brennan, Friedman, & Pollard, 1987; Gordon, Grosz, & Gilliom, 1993; Grosz, Joshi, & Weinstein, 1995; Animacy: Bock, 1982; Bock & Warren, 1985;Branigan, Pickering, & Tanaka, 2008; Rosenbach, 2008, Semantic Richness: Hofmeister, 2011; Karimi & Ferreira, 2016a; Troyer et al., 2016), these results suggest that NPs that are more accessible in memory tend to be later realized with pronouns rather than repeated nouns. However, the potential effect of linear placement of the semantic modification has not yet been examined. The prediction from these findings for the present experiment is straightforward: If modifier position affects the accessibility of associated NPs, we should observe reliable variations in pronominal reference to pre- vs post-modified NPs, with the modification type that results in more accessibility, leading to more pronominal reference to the associated NP. To test this hypothesis, we created experimental stimuli such as (1) in which one of two NPs was either pre-modified (1b-c), post-modified (1d-e), or unmodified, which constituted the baseline condition (1a).
(1)
| Modifier Position | Order of NPs | Example | |
|---|---|---|---|
| a) | Baseline | Baseline | The king criticized the prince. |
| b) | Pre-modified | Target-Competitor | The cruel and merciless kingtarget criticized the princecompetitor. |
| c) | Pre-modified | Competitor-Target | The kingcompetitor criticized the cruel and merciless princetarget. |
| d) | Post-modified | Target-Competitor | The kingtarget who was cruel and merciless criticized the princecompetitor. |
| e) | Post-modified | Competitor-Target | The kingcompetitor criticized the princetarget who was cruel and merciless. |
In all experimental sentences, one NP was always mentioned first (i.e., NP1, king), and the other NP was always mentioned second (i.e., NP2, prince). In the baseline condition (1a), the two NPs were unmodified. In the other four conditions (1b–e), either NP1 or NP2 was either pre- or post-modified. Since it is possible that semantic richness on the part of the modified NP (king) might not only increase activation of that NP but also decrease activation for the co-present unmodified NP (prince, see Baddeley, 1986; Just & Carpenter, 1992; Jäger et al., 2017; Stanford & Garrod, 1981; cf. Foraker & McElree, 2007), we categorized the NPs in the following way: In each condition, we called the modified NP the “target” and the other NP the “competitor”. Targets and competitors are explicitly indicated by subscripts in (1). With this grouping, targets and competitors could be either NP1 or NP2. This design also allowed us to collapse across targets and competitors within each of the pre- and post-modified conditions, reducing the number of NPs to compare and maximizing the power of our analyses (see below). Also, because neither NP is modified in the baseline condition, there are no targets or competitors in this condition. In the two pre-modified conditions, the targets were modified by preceding adjectives, and in the two post-modified conditions, the targets were modified by the same adjectives, but using a relative clause. As is clear from (1), the modifying adjectives were constant across all relevant conditions.
Participants
Sixty undergraduate students from the participant pool of the University of California, Davis took part in this experiment in exchange for course credit. They were all native speakers of American English and reported no language-related disorders.
Stimuli
Forty experimental sentences such as (1) were created. NP1 (king) was always also the syntactic subject and NP2 (prince) always assumed the syntactic object role. Five versions were created for each experimental item, as illustrated in (1). The gender of the two NPs was always the same within an experimental sentence. We did this because our previous work showed near-ceiling pronominal reference when the genders of the two NPs were different (Karimi et al., 2014). Thus, keeping the genders the same was intended to create enough variation in the forms of referring expressions produced by participants to allow us to test our hypotheses. Five experimental lists were created so that each list contained only one version of an experimental item. We also made 60 fillers that were interspersed with the experimental items in a pseudo-randomized order. That is, all the sentences (filler and experimental sentences) were randomized once and all lists were presented in that same order for all participants. The full list of experimental sentences for this experiment is provided in Appendix 10.
Procedure
Participants received each experimental list in the form of a booklet and were asked to write a continuation for each sentence. There were no restrictions on how to produce the continuations and the participants were free to say whatever they wished. The participants were encouraged to produce their continuations “quickly” and “with the first thing that comes to mind”, but there was no time limit for the experiment. Participants were allowed to take a break whenever they wished, but they were instructed not to look at the pages ahead during the break. The experimental session lasted about 45 minutes.
Coding
Because syntactic subjects are considerably more likely to be referred to with pronouns relative to syntactic objects (see above), we first coded whether the participants talked about NP1 (i.e., the syntactic subject) or NP2 (the syntactic object) in their responses. Then we coded the main dependent variable, namely, how the participants referred to either NP1 or NP2 – with a pronoun or with a repeated noun. Note that coding for the preference to talk about NP1 vs. NP2 was critical because any variations in pronominal reference could arise from this preference rather than our manipulations.
Responses were excluded if (1) the referring expression referred to neither the target nor the competitor, (2) collective reference (e.g., They, or, The king and the prince) was used4 (3) neither a pronoun nor a repeated noun was used as a referring expression, (4) the referring expression was not the first-mentioned entity in the response, (5) participants did not produce a new sentence, and (6) the referring expression was part of a subordinate clause in the response (e.g., When he/the prince left, he/the king started contemplating). The application of these criteria resulted in the loss of 502 responses (20.9%)5. Moreover, because the two NPs were of the same gender, the responses employing pronouns could sometimes be ambiguous. It is important to mention, however, that the majority of responses were unambiguous even when a pronoun was used in the response. This is because the NP that was not the referent of the pronoun was usually mentioned in the response at a later point, revealing the referent of the pronoun. An example would be: “He never trusted the prince”. We had 250 truly ambiguous responses in total. To maintain objectivity, these responses were coded by a research assistant who was blind to the purpose of the experiment. To ensure that our results are not biased by ambiguous responses, we repeated the analyses for this experiment (see below) excluding all the 250 ambiguous responses, but the main results stayed the same. The raw frequencies and percentages of references to NP1, NP2 and excluded responses are reported in Appendix 1.
Statistical Analyses
We coded whether reference to targets or competitors was accomplished through a pronoun (he) or a repeated noun (say, the king), creating a binomial measure. In other words, we collapsed over the two NPs of the same type (i.e., target, or competitor) within each of pre- and post-modified conditions. For the baseline condition, we simply collapsed over the two NPs. Thus, our comparisons always involved comparing the probability of pronominal reference for the following NPs: Pre-modified targets (i.e., king in 1b, and prince in 1c), pre-modified competitors (i.e., prince in 1b, and king in 1c), post-modified targets (i.e., king in 1d, and prince in 1e), post-modified competitors (i.e., prince in 1d, and king in 1e), and the two NPs in the baseline condition (i.e., king and prince in 1a).
It is also important to note that the NPs that were compared always had the same syntactic roles and linear positions. For instance, a comparison between overall pronominal reference to the two NPs in the baseline condition with pre-modified targets, would be comparing pronominal reference to king and prince in (1a) with king and prince in (1b) and (1c), respectively. Critically, whenever we observed a reliable difference in rate of pronominal reference for a particular comparison across conditions, we also report the preference to talk about NP1 (vs. NP2) for that same comparison to see if the results could be reduced to a preference to talk about NP1 (rather than modifier position).
To examine the full effects of our manipulations on forms of reference, we conducted two separate analyses. In the “Modification Type Analysis”, we removed the two competitors, creating a 3-level predictor: Pre-modified target, Post-modified target, and Baseline. We then tested whether the probability of pronominal reference differs between pre- and post-modified targets relative to the baseline condition, and also between pre- and post-modified targets. In the “Relative Richness Analysis”, we removed the Baseline condition, creating 2 predictors with 2 levels each: Modifier-Position (Pre-modified vs. Post-modified), and NP Type (Target vs. Competitor), testing whether the relative semantic richness effect between targets and competitors, if any, varies across Pre- and Post-modified conditions. In other words, this analysis tested the critical interaction between Modifier Position and NP Type; does the difference in pronominal reference to targets and competitors vary across the levels of modifier position?
Our analyses incorporated a multi-level modeling approach. Following Barr et al. (2013), we attempted to keep the random-effects structure as “maximal” as possible. However, because most of our models with “full” random-effects structures failed to converge, we had to simplify the models. Specifically, for the models reported in this paper, we consistently ran “intercept-only” models for the Modification Type Analysis, but for the “Relative Richness Analysis”, because the interaction term was the critical effect, we always included by-subject and by-items random slopes for the interaction term, but we did not include random intercepts.
Results
Figure 1 displays the percentage of pronominal reference (out of pronominal plus repeated noun referring expressions) in each condition, and Table 1 reports the results of our statistical analyses. As is clear from this table, in the Modification Type Analysis, we observed significantly more pronominal reference to post-modified targets than to unmodified NPs in the baseline condition. However, the probability of pronominal reference did not differ between pre-modified targets relative to the baseline condition. There was also a trend towards more pronominal references to post- than to pre-modified NPs. In the Relative Richness Analysis, we observed a main effect of NP Type, no main effect of Modifier Position and, critically, a reliable interaction between these two factors. Follow-up simple effects revealed a significant difference between the target and the competitor within the post-modified condition, but not within the pre-modified condition. The raw frequencies and percentages of pronouns and repeated nouns are provided in Appendix 2.
Figure 1.
Probability of pronominal reference in each Condition. Experiment 1.
Table 1.
Results of analyses on probability of pronominal reference. Experiment 1. “PreM” and “PostM” stand for Pre-modified and Post-modified, respectively.
| Analysis | Contrast | β | SE | z | p |
|---|---|---|---|---|---|
| Modification Type | PreM_Target vs. Baseline | .08 | .20 | .42 | .67 |
| PostM_Target vs. Baseline | .45 | .21 | 2.15 | .03 | |
| PreM_Target vs. PostM_Target | .36 | .21 | 1.75 | .08 | |
| Relative Richness | Modifier Position | −.01 | .11 | −.17 | .86 |
| NP Type | .45 | .11 | 4.07 | <.001 | |
| Modifier Position ×NP Type | .47 | .22 | 2.13 | .03 | |
| NP Type within PreM | −.14 | .22 | −.64 | .52 | |
| NP Type within PostM | −.58 | .26 | −2.20 | .03 |
Importantly, there were no significant differences in how often the participants talked about NP1 (vs. NP2) in Modification Type or in Relative Richness analyses. The full results for the preference to talk about NP1 vs. NP2 are provided in Appendix 3.
Discussion
The results of Experiment 1 provided some evidence that post-modifiers might lead to greater memory activation of the associated NPs than pre-modifiers do: First, while the probability of pronominal reference did not differ between the pre-modified targets and the unmodified NPs in the Baseline condition, there was reliably more pronominal references to post-modified NPs relative to unmodified NPs in the Baseline condition. Second, there was a trend towards more pronominal references to post- than to pre-modified NPs. And third, we observed more pronominal reference to post-modified targets than to co-present competitors, but no difference between pre-modified targets and co-present competitors. Interestingly, this higher memory activation for post-modified targets seems not only to increase pronominal reference to the targets themselves, but also to decrease pronominal reference to the competitors (compare the blue bars with the gray bar in Figure 1). Note that we also observed a trend towards more pronoun use for post-modified NPs compared to the unmodified NPs in the baseline condition. However, because there was also a trend towards talking about NP1 more for post-modified targets than in the baseline condition, this result is difficult to interpret and we will not discuss it further. To the best of our knowledge, these results are the first to show that pre- and post-modifiers result in variations in how easily the associated NPs are retrieved from memory at a subsequent point. We will discuss the theoretical implications of these observations in the discussion section of a meta-analysis that combines the data from the first three experiments (see below).
Experiment 2
While the results of Experiment 1 showed enhanced subsequent retrieval for post-modified NPs, the participants were able to re-read the given sentences (theoretically, an endless number of times), which might have mitigated the effect of memory decay caused by the temporal distance between the NPs in the critical sentences and the referring expressions in the responses. Specifically, post-nominal relative clauses necessarily increased the distance between a post-modified head noun and when reference production was initiated, and yet they received more pronominal reference than pre-modified NPs did. However, the greater memory activation for post-modified NPs might have originated from more re-reading of post-modified NPs compared to pre-modified NPs to compensate for the greater temporal (and linear) distance between the head noun of post-modified NPs and the end of the given sentence. Removing the possibility of re-reading would help rule out this alternative explanation. In order to control for re-reading, we conducted a second experiment in the spoken modality in which participants heard the critical sentences (only once) and spoke their continuations (instead of writing them down).
Participants
Seventy undergraduate students from the participant pool of University of California, Davis took part in this experiment in exchange for course credit. They were all native speakers of American English and reported no language-related disorders.
Stimuli
The experimental sentences were very similar to those used in Experiment 1. The only difference was that one of the NPs was replaced with another NP of a different gender, as illustrated in (2). We made the genders of the critical NPs different in this experiment because a former experiment of ours showed overall less pronoun use in the spoken than in the written modality (Karimi et al., 2014). We also know from previous research that same-gender NPs result in less pronoun use (Arnold & Griffin, 2007). Thus, keeping genders the same and presenting the sentences auditorily could have resulted in very low pronominal reference rates, obscuring any effects. The full list of experimental sentences for this experiment is provided in Appendix 10.
(2)
| Modifier Position | Order of NPs | Example | |
|---|---|---|---|
| a) | Baseline | Baseline | The king criticized the princess. |
| b) | Pre-modified | Target-Competitor | The cruel and merciless kingtarget criticized the princesscompetitor. |
| c) | Pre-modified | Competitor-Target | The kingcompetitor criticized the cruel and merciless princesstarget. |
| d) | Post-modified | Target-Competitor | The kingtarget who was cruel and merciless criticized the princesscompetitor. |
| e) | Post-modified | Competitor-Target | The kingcompetitor criticized the princesstarget who was cruel and merciless. |
Procedure
The instructions were identical to those used in Experiment 1. The experiment was programmed in PsychoPy2 (v1.83.03). In each experimental trial, the participants pressed the spacebar to hear the prompt sentence. After the presentation of the current sentence was over, a “speak” prompt would appear on the center of the computer screen, indicating that the participants could start speaking their responses. The responses were recorded by an in-built microphone during the time the participants were speaking. Since the length of the responses was expected to vary for each item and participant, we programmed the experiment in a way that the recording of the responses was controlled by the participants themselves; pressing the spacebar during the time the “speak” prompt would terminate recording, and display the prompt for the next trial (i.e., “press the spacebar to hear the next sentence”).
Coding and statistical analyses
The exclusion criteria was identical to that in Experiment 1, except that there were no ambiguous responses. The application of the exclusion criteria resulted in the loss of 1077 (38.4%) of data6. The raw frequencies and percentages of references to NP1, NP2 and excluded responses are provided in Appendix 4. The statistical analyses were identical to those in Experiment 1.
Results
Figure 2 shows the percentage of pronominal reference (i.e., pronouns out of pronouns plus repeated noun referring expressions) in each condition, and Table 2 reports the results of our statistical analyses. As is clear from this table, in the Modification Type Analysis, we observed reliably more pronominal reference to both the pre- and post-modified targets relative to the Baseline condition. However, although the there was a numerical trend towards more pronominal reference for post- than pre-modified NPs, this effect did not reach statistical significance. In the Relative Richness Analysis, we observed a main NP Type effect, with reliably more pronominal reference to targets that to competitors, but no main effect of Modifier Position. Moreover, the critical interaction between Modifier Position did not reach statistical significance. The raw frequencies and percentages of pronouns and repeated nouns are reported in Appendix 5.
Figure 2.
Probability of pronominal reference for each Condition. Experiment 2.
Table 2.
Results of analyses on probability of pronominal reference. Experiment 2. “PreM” and “PostM” stand for Pre-modified and Post-modified, respectively.
| Analysis | Contrast | β | SE | z | p |
|---|---|---|---|---|---|
| Modification Type | PreM_Target vs. Baseline | .54 | .21 | 2.59 | .009 |
| PostM_Target vs. Baseline | .84 | .22 | 3.74 | <.001 | |
| PreM_Target vs. PostM_Target | .30 | .21 | 1.38 | .16 | |
| Relative Richness | Modifier Position | −.01 | .10 | −.11 | .90 |
| NP Type | .35 | .10 | 3.29 | <.001 | |
| Modifier Position ×NP Type | .31 | .21 | 1.44 | .14 | |
| NP Type within PreM | −.18 | .27 | −.66 | .50 | |
| NP Type within PostM | −.72 | .27 | −2.6 | .007 |
Importantly, talking about NP1 was significantly more likely for pre-modified targets relative to the Baseline condition (β = .39, SE= .17, Z= 2.28, p = .02). However, the probability of talking about NP1 did not reliably differ between post-modified targets and the Baseline condition (β = .19, SE= .18, Z= −1.05, p = .29). The full results for NP1 vs. NP2 reference are provided in Appendix 6.
Although the interaction between NP Type and Modifier Position did not reach statistical significance, there was justification to examine the simple effects: First, this interaction was reliable in Experiment 1, 2. There was a numerical trend in the same direction in the current experiment (p = .14), and 3. The tendency to talk about NP1 was greater for pre-modified targets than in the Baseline condition while this tendency did not reliably vary between post-modified targets and the Baseline condition. This means that pre-modified targets were more likely than post-modified targets to be subsequently realized with pronouns due to greater NP1 reference, making it difficult for the interaction to reach statistical significance. As shown in Table 2 and consistent with the results of Experiment 1, an analysis of simple effects revealed that the probability of pronominal reference was reliably greater for the target relative to the competitor within the Post-modified condition but not within the Pre-modified condition.
Discussion
Consistent with the results of Experiment 1, the results of Experiment 2 also showed a reliably stronger tendency to use pronouns (rather than repeated NPs) to refer to post- than to pre-modified NPs: The probability of pronominal reference was reliably greater for targets relative to competitors within the post-modified condition, but not within the pre-modified condition, suggesting that the retrieval of a post-modified NP’s representation from memory is relatively easier compared to that of a pre-modified (and an unmodified) NP. Importantly, because the critical sentences were presented auditorily (and only once), the greater activation of post-modifiers could not be due to re-reading of the post-modified NPs. Thus, the results of Experiment 2 lend even stronger support to the observation that post-modifiers lead to greater memory activation of the head noun than pre-modifiers. We will discuss the theoretical implications of these results after we present a meta-analysis combining the data from the first three experiments (see below).
In addition, unlike in Experiment 1 in which pre-modifiers not only increased pronominal reference to the target but also decreased pronoun use for the competitor, in Experiment 2 post-modifiers only increased pronominal reference to the target and did not affect the rate of pronoun use for the competitor (compare the blue bars with the gray bar in Figure 2). This pattern of results is consistent with our previous findings and seems to be related to the stimulus modality (spoken vs. written, see Karimi et al., 2014). However, since this observation is tangential to our research questions, we will not discuss it any further.
Experiment 3
In the first and second experiments we observed more pronominal reference to post- than to pre-modified NPs, suggesting that post-modified NPs are more highly activated in memory than pre-modified NPs are. However, we did not find clear evidence for higher activation of pre-modified NPs compared to unmodified NPs, which is inconsistent with previous findings (e.g., Hofmeister, 2011). In an attempt to obtain even clearer results for both pre- and post-modified NPs, we made all NP1s inanimate in Experiment 3 in order to reduce the overall probability of talking about NP1 (see Fukumura & Van Gompel, 2011). Since NP1s are already highly activated, we hoped that more reference to NP2 might lead to clearer results of modifier position.
Participants
Sixty undergraduate students from the participant pool of University of California, Davis took part in this experiment in exchange for course credit. The data from one participant was removed from the analysis because they only produced repeated nouns (and no pronouns) in their responses. All participants were native speakers of American English and reported no language-related disorders.
Stimuli
We created 50 experimental items for this experiment. The sentences were very similar to those used in Experiments 1 and 2. The only difference was that NP1 was made inanimate across all the conditions, and as a result of this, the modifying information for animate and inanimate NPs were different, as is shown in (3). The full list of experimental sentences for this experiment is provided in Appendix 11.
(3)
| Modifier Position | Order of NPs | Example | |
|---|---|---|---|
| a) | Baseline | Baseline | The carpet mesmerized the photographer for hours in the museum. |
| b) | Pre-modified | Target-Competitor | The ancienttarget and ornate carpet mesmerized the photographercompetitor for hours in the museum. |
| c) | Pre-modified | Competitor-Target | The carpetcompetitor mesmerized the stylish and artistic photographertarget for hours in the museum. |
| d) | Post-modified | Target-Competitor | The carpettarget that was ancient and ornate mesmerized the photographercompetitor for hours in the museum. |
| e) | Post-modified | Competitor-Target | The carpettarget mesmerized the photographercompetitor who was stylish and artistic for hours in the museum. |
Procedure, coding and statistical analyses
The procedure was identical to that in Experiment 2, and the coding and statistical analyses were identical to those in Experiments 1 and 2. Note that the pronouns were always unambiguous in this Experiment; “it” for the NP1s and “he” or “she” for NP2s. The application of the exclusion criteria resulted in the loss of 1144 (38.8%) responses7. The raw frequencies and percentages of references to NP1, NP2 and excluded responses are reported in Appendix 7. As in the first two experiments, either the “full”, or the “no-correlation”, or the “slopes-only” models converged (see above and Barr et al., 2013) for all of our analyses, and we therefore did not need to simplify the random effects structures of our models any further.
Results
The animacy manipulation was successful: Participants talked about NP2 (the animate NP) much more than NP1 (the animate NP) across all conditions (NP1 reference = 381, NP2 reference = 1425), which is consistent with previous research (Fukumura & Van Gompel, 2011). The percentage of pronominal reference (i.e., pronouns out of pronouns plus repeated noun referring expressions) for each condition is shown in Figure 3, and the results of our statistical analyses are reported in Table 3. As can be seen in this table, in the Modification Type Analysis, we observed reliably more pronominal reference to post-modified targets relative to the baseline condition. The probability of pronominal reference was also greater for pre-modified targets relative to the Baseline condition, but this effect was marginally significant. The probability of pronominal reference did not differ between the pre- and post-modified conditions. Consistent with Experiments 1 and 2, in the Relative Richness Analysis, the Modifier Position effect was not significant, but the effect of NP Type was statistically reliable, with greater pronominal reference to targets that to competitors. However, similar to Experiment 2, the critical interaction between Modifier Position and NP Type did not reach statistical significance. The raw frequencies and percentages of pronouns and repeated nouns can be found in Appendix 8.
Figure 3.
Probability of pronominal reference for each Condition. Experiment 3.
Table 3.
Results of analyses on probability of pronominal reference. Experiment 3. “PreM” and “PostM” stand for Pre-modified and Post-modified, respectively.
| Analysis | Contrast | β | SE | z | p |
|---|---|---|---|---|---|
| Modification Type | PreM_Target vs. Baseline | .42 | .22 | 1.92 | .05 |
| PostM_Target vs. Baseline | .62 | .22 | 2.76 | .005 | |
| PreM_Target vs. PostM_Target | .19 | .22 | .87 | .38 | |
| Relative Richness | Modifier Position | −.01 | .11 | −.16 | .87 |
| NP Type | .24 | .11 | 2.15 | .03 | |
| Modifier Position ×NP Type | .17 | .22 | .80 | .42 | |
| NP Type within PreM | −.35 | .33 | −1.06 | .28 | |
| NP Type within PostM | −.90 | .31 | −2.83 | .004 |
Importantly, the rate of talking about NP1 (vs. NP2) was greater for the unmodified NPs in the Baseline condition relative to the pre-modified targets. The full results for NP1 vs. NP2 reference are provided in Appendix 9.
As in Experiment 2, although the main interaction between Modifier Position and NP Type was not significant, we had enough motivation to examine the simple effects because 1. The probability of pronominal reference was reliably greater for post-modified targets relative to the Baseline condition, but only marginally greater for pre-modified targets relative to the Baseline, and 2. The results of the previous two experiments did reveal reliable differences in the simple comparisons. As reported in Table 3, and consistent with Experiments 1 and 2, the probability of pronominal reference was reliably greater for targets relative to the competitors within the post-modified condition but not within the pre-modified condition.
Discussion
Consistent with the results of Experiments 1 and 2, the results of Experiment 3 showed that post-modified NPs are reliably more likely to be subsequently pronominalized than pre-modified NPs, suggesting that the memory representations associated with post-modified NPs are relatively more accessible in memory than those associated the pre-modified NPs. Specifically, and in line with the results of the first two experiments, the probability of pronominal reference was significantly greater for targets than for competitors within the post-modified condition, but not within the pre-modified condition. Moreover, we also observed significantly greater pronominal reference to post-modified targets relative to the baseline condition, whereas pronominal reference was only marginally greater for pre-modified targets relative to the Baseline condition. This pattern of results suggests that post-modified NPs increase the accessibility of the associated memory representation more than pre-modifiers do. In other words, these observations suggest that although both pre- and post-modification of an NP boosts the accessibility of the associated memory representation, post-modifications result in greater activation boosts than pre-modifications. Note that although only marginally significant, the greater pronominal reference to the pre-modified NPs relative to the baseline condition is consistent with previous studies (Hofmeister, 2011). We will discuss the far-reaching theoretical implications of our results below where we present the results of a meta-analysis that combines the data from the first three experiments.
Experiments 1, 2 and 3 combined
In the three experiments reported above, only in Experiment 1 (and not in Experiments 2 and 3) did we observe a greater probability of pronominal reference to post- than to pre- modified targets. Moreover, the critical main interaction between Modifier Position and NP Type in the Relative Richness Analysis was statistically reliable only in Experiment 1. As is well-known, the probability of Type 1 error rates is higher if simple effects are tested directly without a main interaction. However, since we consistently observed greater pronominal reference rates to targets than to competitors within the post-modified condition but not within the pre-modified condition across all three experiments, we believe the reason the main interaction did not reach statistical significance in Experiments 2 and 3 is simply lack of statistical power. It is important to note that both of the targets were semantically rich and therefore resulted in numerically more pronominal reference relative to the competitors, making it more difficult for the main interaction to reach statistical significance. However, a direct comparison between pre- and post-modified targets as well as testing the main interaction between Modifier Position and NP Type are critical for our claims. This is because if post-modifiers do result in higher memory activation of the head noun than pre-modifiers, there should be more pronominal reference to post- than pre-modified targets, and the difference in pronominal reference between targets and competitors should be greater within the post-modified condition relative to the pre-modified condition.
To maximize statistical power, we combined the data from all three experiments reported above and repeated the Modification Type and Relative Richness Analyses. Table 4 reports the results of these analyses and Figure 4 displays the proportion of pronouns (out of pronouns plus repeated nouns) for all conditions.
Table 4.
Results of analyses on probability of pronominal reference. Experiments 1, 2 and 3 combined. “PreM” and “PostM” stand for Pre-modified and Post-modified, respectively.
| Analysis | Contrast | β | SE | z | p |
|---|---|---|---|---|---|
| Modification Type | PreM_Target vs. Baseline | .34 | .12 | 2.82 | .004 |
| PostM_Target vs. Baseline | .63 | .12 | 4.99 | <.001 | |
| PreM_Target vs. PostM_Target | .29 | .12 | 2.31 | .02 | |
| Relative Richness | Modifier Position | −.00 | .06 | −.07 | .93 |
| NP Type | .32 | .06 | 5.23 | <.001 | |
| Modifier Position ×NP Type | .31 | .12 | 2.51 | .01 | |
| NP Type within PreM | −.18 | .12 | −1.48 | .13 | |
| NP Type within PostM | −.75 | .14 | −5.32 | <.001 |
Figure 4.
Probability of pronominal reference for each Condition. Experiments 1, 2 and 3 combined.
As can be seen in Table 4, the Modification Type analysis showed that the probability of pronominal reference was significantly greater for both pre- and post-modified targets than to the unmodified NPs in the Baseline condition. Critically, however, this analysis also revealed that post-modified targets were significantly more likely to be subsequently realized with pronouns than pre-modified targets were. Moreover, the Relative Richness Analysis showed a significant main interaction between Modifier Position and NP Type such that the probability of pronominal reference was reliably greater for targets relative to competitors within the post-modified condition but not within the pre-modified condition. Thus, the results from our combined analysis lends direct support to the finding that post-modified NPs are more accessible in memory than are pre-modified NPs. We observed no reliable differences in the probability of talking about NP1 vs. NP2 in the Modification Type or Relative Richness analyses.
Discussion
The analysis of the combined data from the first three experiment revealed a clear advantage for post-modified NPs relative to pre-modified NPs in terms of probability of pronominal reference: Post-modified NPs were significantly more likely to be subsequently realized with pronouns than were pre-modified NPs, and targets were reliably more likely to be pronominalized than competitors within the post-modified condition but not within the pre-modified condition.
To the best of our knowledge, our results are the first to demonstrate that the position of modifiers causes variations in ease of subsequent retrieval. The greater accessibility of post- vs pre-modified NPs could be caused by a greater perceptual significance attached to post- than pre-modifiers, or by the fact they are consistent with a given-before-new information structure, or by the smaller memory demands associated with processing post-modifiers (see Introduction).
In any case, because the modifying adjectives were constant across the pre- and post-modified NPs, these results indicate that distinctiveness is not the only cause for retrieval ease of semantically richer NPs, and that other perceptual, information structure or memory factors have a role to play. Moreover, the results clearly show that memory decay does not negatively impact the retrieval of post-modified NPs despite the fact that they increased the temporal distance between the encoding of the retrieval of the head noun, which is consistent with previous findings (Karimi & Ferreira, 2016a; Karimi, Swaab, & Ferreira, 2018). Finally, the fact the post-modified NPs led to more pronominal reference than did pre-modifiers clearly suggests that predictability of the head noun category does not play a key role during the encoding of semantically rich words: Although the syntactic category of “noun” could be predicted in the case of pre-modified NPs, but not in the case of post-modified NPs, we actually observed more pronominal reference to post- than to pre-modified NPs.
Experiment 4
Experiments 1, 2 and 3 assessed the effect of modifier position on subsequent retrieval during language production, and pronominal reference was used as a measure of retrieval ease. However, some previous studies have investigated the semantic richness effect during online language comprehension, using reading times as a measure of memory activation and ease of retrieval (Hofmeister, 2011; Troyer et al., 2016). In order to assess whether our results from language production also apply to language comprehension using a more conventional measure of memory activation/retrieval, we conducted a self-paced reading experiment using experimental stimuli such as (4) and (5). To simplify the design, we also removed the competitor form the experimental sentences.
(4) Sentence 1
a) Baseline: The king was diagnosed with a terminal disease.
b) Pre-Modified: The cruel and merciless king was diagnosed with a terminal disease.
c) Post-Modified: The king who was cruel and merciless was diagnosed with a terminal disease.
(5) Sentence 2
For almost two months, he had no idea because no one would dare break the news.
Previous research has demonstrated that pronouns trigger the retrieval of the representations associated with their referents (Dell, McKoon, & Ratcliff, 1983; Gernsbacher, 1989; Gerrig & McKoon, 1998; Lucas, Tanenhaus, & Carlson, 1990; MacDonald & MacWhinney, 1990; Sanford & Garrod, 1989, 2005), and that the ease of that retrieval directly influences the ease of processing the pronoun (Karimi et al., 2018). As such, if the results from the first three experiments generalize to language comprehension, we should observe faster reading times on the pronoun (and/or the subsequent regions) following post-modified than following pre-modified NPs.
Participants
Sixty undergraduate students from the participant pool of University of California, Davis took part in the experiment in exchange for course credit. We removed the data from one participant because they did not complete the experiment. All participants were native speakers of American English and reported no language-related disorders.
Stimuli
We created 33 experimental discourse segments such as (4) and (5). Each sentence contained one human NP (the king) which always assumed the syntactic subject role, and was either unmodified, pre-modified or post-modified, creating the 3 conditions of interest, as illustrated in (4). All three versions of an experimental item were followed by the same second sentence that contained a pronoun (either he or she) referring to the NP in the first sentence. The full list of stimuli for this experiment is provided in Appendix 12. The experimental sentences were intermixed with 40 other fillers that in fact served as experimental stimuli for another independent study which did not investigate pronoun processing. The 33 experimental sentences and the 40 fillers were distributed among three lists such that each participant was exposed to only one version of each experimental sentence. The order of the sentences in each list was randomized for each participant such that each participant viewed each list in a different order. Twelve experimental sentences and 20 fillers were tagged with a True/False comprehension question to ensure that participants pay close attention to the task. The experiment was programmed in PsychoPy2 (v1.83.03) which presented the stimuli and recorded reaction times for each button press (see below).
Procedure
The experiment was conducted individually and in a quiet room. First, the instructions appeared on the screen which stated that the participants were about to read two-sentence stories, and that their job is to read them for comprehension. The instructions also stated that a comprehension question in the form of a True/False statement would appear for a random number of stories. An experimental trial started with participants pressing the space bar on the keyboard, which led to the display of the first sentence on the computer screen all at once. When the participant was done reading the first sentence, they pressed the spacebar again to trigger the presentation of the second sentence. Unlike the first sentence, the second sentence was presented one word at a time, each word appearing with the press of the spacebar and staying on the screen until the spacebar was pressed again, which triggered the presentation of the next word, and so on and so forth until the second sentence ended. If the current item had a comprehension question, it would appear on the screen (in full) after the last word of the second sentence was read, and the participant had to press “1” on the Num Lock pad to indicate True and “2” to indicate false. If there was no question for the current trial, the screen would show “No question” and the participants had to press 0 on the Num Lock to make it disappear, and press spacebar to move on to the next item.
Statistical Analyses
We performed linear mixed effects regression models on the data, with reading time as the dependent variable, and Modifier Position as the predictor. We Helmert coded the predictor to test both the general effect of modifications (Baseline vs. the average of Pre-Modified and Post-Modified), as well as the direct contrast between the two modifiers (Pre-modified vs. Post-modified). The protocol for determining the random-effects structure was identical to that in the first three experiments. However, for this experiment the “full” models (involving random intercepts for both subjects and items as well as by-subjects and by-items random slopes for the effect of Modifier Position) always converged and we did not need to simplify the random structure of our models any further (Barr et al., 2013). We considered any t values greater than 2 as statistically significant (Gelman & Hill, 2006).
Results
Prior to analyzing the data, we log-transformed the reading times to approach a normal distribution of reading times. We then removed reaction times that fell 2.5 standard deviations below or above the overall mean (425, or 1.7% of the total trials). We also noticed that a few of the first sentences had unreasonably fast or slow reading times. Since the manipulations occurred in the first sentence and a failure to read those sentences would result in no exposure to the manipulation, we also removed trials where the reading time of the first sentence was too long or too short (29, or .001% trials). Unreasonable reading times on Sentence 1 were defined in the following way: We first calculated the fastest reasonable reading time as 2.5 SDs below the mean reading time of the shortest Sentence 1, and the longest reasonable reading time as 2.5 SDs above the mean of the longest Sentence 1. We then removed any reading times that were faster than the fastest reasonable reading time or slower than the longest reasonable reading time.
Figure 5 displays the reading times for the four words preceding and the five words following the critical pronoun (represented by “n”) in Sentence 2.
Figure 5.
Reading time by region and Modifier Position. Experiment 4. “n” represents the critical pronoun.
Table 5 reports the results of our statistical analyses. As is clear from this table, there was a general effect of modification in the two regions immediately following the critical pronoun (“n+1” and “n+2”), with reading times in the average of Pre- and Post-modified conditions being faster than the Baseline condition. Importantly, on the fourth word following the critical pronoun (“n+4”), we also observed faster reading times for the Post-modified condition relative the Pre-modified condition.
Table 5.
Results for the statistical analyses for Experiment 4.
| Region | Contrast | Estimate | SE | t |
|---|---|---|---|---|
| n-4 | Baseline vs. (PreM & PostM) | −0.00 | .01 | −.06 |
| PreM vs. PostM | −0.00 | .01 | −.00 | |
| n-3 | Baseline vs. (PreM & PostM) | .02 | .01 | 1.72 |
| PreM vs. PostM | −.00 | .01 | 0.40 | |
| n-2 | Baseline vs. (PreM & PostM) | .02 | .01 | 1.86 |
| PreM vs. PostM | −.00 | .01 | −.26 | |
| n-1 | Baseline vs. (PreM & PostM) | .02 | .01 | 1.64 |
| PreM vs. PostM | .01 | .01 | 1.02 | |
| n | Baseline vs. (PreM & PostM) | .01 | .01 | 1.02 |
| PreM vs. PostM | .00 | .01 | .29 | |
| n+1 | Baseline vs. (PreM & PostM) | .03 | .01 | 2.39 |
| PreM vs. PostM | −.00 | .01 | −.03 | |
| n+2 | Baseline vs. (PreM & PostM) | .02 | .01 | 2.01 |
| PreM vs. PostM | .01 | .01 | 1.13 | |
| n+3 | Baseline vs. (PreM & PostM) | .02 | .01 | 1.67 |
| PreM vs. PostM | .01 | .01 | .94 | |
| n+4 | Baseline vs. (PreM & PostM) | .01 | .01 | 1.21 |
| PreM vs. PostM | .03 | .01 | 2.23 | |
| n+5 | Baseline vs. (PreM & PostM) | .02 | .01 | 1.53 |
| PreM vs. PostM | .00 | .01 | .25 |
Accuracy for answering the comprehension questions was 91.1% in the Baseline condition, 87.2% in the Pre-modified condition, and 87.8% in the Post-modified condition respectively. The differences in accuracy rates were not statistically reliable between any of the conditions (Pre-modified vs. Baseline: p = .21, Post-modified vs. Baseline: p = .25, Pre-modified vs. Post-modified: p = .92).
Discussion
Consistent with the results of the production experiments, the results of the comprehension experiment showed easier retrieval for post-modified NPs relative to pre-modified (and unmodified) NPs. Specifically, reading times for fourth word following the critical pronoun was faster for post-modified than for unmodified NPs. Thus, the results of Experiment 4 provide further evidence that the choice between pre- and post-modifiers results in measurable processing cost differences, which might be due to the greater perceptual significance associated with post-modifiers, or the given-before-new structure of post-modified NPs, or a maintenance cost associated with processing pre-modified NPs. Moreover, and consistent with the results of the previous three experiments, the results of the current experiment demonstrate that distinctiveness is not sufficient for explaining why semantically richer NPs are subsequently retrieved faster from memory, that the effect of time-based decay is overridden by post-modifying information, and also that prediction of the head noun may not facilitate encoding over and above semantic richness. Moreover, the results of Experiment 4 also revealed a general modification effect, with faster reading times for the average of pre- and post-modified NPs relative to unmodified NPs on the two words immediately following the critical pronoun, which is fully consistent with previous findings (Hofmeister, 2011; Karimi & Ferreira, 2016; Karimi et al., 2014; Troyer et al., 2016)
General Discussion
In four experiments, we observed that post-modified NPs are more accessible in memory than are pre-modified NPs. In Experiments 1, we observed significantly more pronominal reference to post-modified NPs than to unmodified competitors, whereas pronominal reference to pre-modified NPs and unmodified competitors did not reliably differ. Moreover, in the first three experiments, we consistently observed a relatively stronger pronominalization tendency for post- than to pre-modified NPs such that post-modified targets were realized with significantly more pronouns than unmodified NPs, whereas pronominalization rates did not differ between pre-modified and unmodified NPs. However, even stronger evidence for the higher memory of post- relative to pre-modified NPs during language production came from the meta-analysis of the first three experiments: In the combined dataset, we observed significantly more pronominal reference to post- than to pre-modified NPs, and also significantly more pronominal reference to targets than to competitors when the target NPs were post-modified, but not when they were pre-modified. Finally, in Experiment 4, we observed faster reading times for pronouns following pre- and post-modified NPs relative to unmodified NPs. However, and critically, we also observed enhanced reading times for pronouns following post-modified compared to pre-modified NPs.
The linear position of modifications is a syntactic choice in English that may or may not have processing consequences. However, although the processing cost associated with alternative syntactic constructions has been extensively studied for many syntactic forms (see the Introduction for a brief review), to the best of our knowledge, the potential effect of modifier position has not been investigated. Thus, our results are the first to show that the linear position of modifiers affects subsequent retrieval of associated memory representations.
Our results are consistent with previous studies showing that semantically richer NPs are more accessible in memory (Hofmeister, 2011; Hofmeister & Vasishth, 2014; Karimi & Ferreira, 2016a; Karimi et al., 2014; Troyer et al., 2016); in all of our experiments, both pre- and post-modified NPs were pronominalized more than unmodified NPs, and were also read faster than unmodified NPs. However, our results contribute to this literature by demonstrating that the way in which semantic richness is conferred (via pre- vs. post-modifiers) also affects the accessibility of the resulting representations. As mentioned above, all three possibilities about the encoding and retrieval of pre- vs. post-modifiers are supported by the current Psycholinguistic literature. One possibility was that compared to post-modifiers, pre-modifiers might be encoded more efficiently in memory because they have the advantage of rendering the syntactic category (plus perhaps some of the semantic features) of the head noun predictable, as well as the advantage of a shorter temporal distance between the head noun and the retrieval point. Another possibility was that there might have been no difference between the encoding efficiency of pre- and post-modifiers, because the modifying adjectives were constant across our critical condition, and therefore the associated representations should be equally distinct in memory, resulting in the same degree of interference during retrieval.
However, our data clearly supported the third possibility, namely, that post-modified NPs are encoded more efficiently in memory than pre-modified NPs. We propose three reasons for this effect: First, post-modifying relative clauses might be more perceptually significant than pre-modifying adjectives. This greater perceptual significance might stem from the fact that post-modifying relative clauses convey more complex information. For instance, unlike pre-modifying adjectives, post-modifying relative clauses can include whole clauses (“The king who was sitting on the throne”), additional NPs (“The king who kissed the queen”), more specific information (“The king who was cruel in the past”), among other types of complex information. Such functional distribution of pre- vs. post-modifiers are perhaps learned by comprehenders, causing them to devote more attentional and/or memory resources to post- than to pre-modifiers, which in turn may lead to more robust representations for post-modified NPs. Note that this explanation is consistent with previous research showing that comprehenders are sensitive to the significance of subordinate vs. main clauses. For example, past research has demonstrated that comprehenders are significantly less likely to detect false assertions when they occur in subordinate clauses than in main clauses (Baker & Wagner, 1987), and that subordinate clauses are more likely to be processed in a shallow manner than main clauses (Christianson, Hollingworth, Halliwell, & Ferreira, 2001; Christianson, Williams, Zacks, & Ferreira, 2006; Ferreira, Christianson, & Hollingworth, 2001; also see Ferreira, Bailey & Ferraro, 2002; Karimi & Ferreira, 2016b; Sanford & Sturt, 2002). Similarly, it has been argued that prediction of upcoming information is less likely to occur in subordinate clauses compared to main clauses (Ferreira & Lowder, 2016). Clearly, the perceptual significance explanation for the greater accessibility of post- vs. pre-modified NPs is also consistent with previous research showing that comrehenders are sensitive to statistical regularities of linguistic input, and can learn from these regularizes to facilitate future processing (Chang, Dell, Bock, & Griffin, 2000; Chang, Dell, & Bock, 2006; Conway, Bauernschmidt, Huang, & Pisoni, 2010; Jaeger, 2010; Clark, 2013; Kleinschmidt, Fine, & Jaeger, 2012; Levy, 2008; MacDonald, 2016; McDonald & Shillcock, 2003).
Second, post-modified NPs are consistent with the given-before-new information structure, and this structure has been shown to facilitate processing (Ariel, 1990, 1996; Benatar & Clifton, 2014; Chafe, 1976; Clark & Clark, 1978; Halliday, 1967; Haviland & Clark, 1974). Since the head noun functions as the core concept to which the modifications are added, the head noun of modified NPs may constitute given information, whereas the modifications may constitute new information. Critically, because modifying information follows the head noun in the case of post-modified NPs, but precedes it in the case of pre-modified NPs, post-modified NPs are consistent with the given-before-new format of how information is packaged, whereas pre-modified NPs are not. As such, post-modified NPs might be easier to process and therefore easier to encode, leading to facilitated subsequent retrieval. An important point pertaining to the information structure of pre- vs. post-modified NPs also merits discussion here: Under some theories of language processing, ease of sentence comprehension is largely determined by preferences during language production, with constructions preferred and generated more frequently during production being easier to comprehend (MacDonald, 2013; 2016; Hopman & MacDonald, 2018, also see Levy, 2008). Since production proceeds in an incremental manner (Bock, 1982; Bock & Warren, 1985; De Smedt & Kempen, 1987; Kempen & Hoenkamp, 1987; Levelt, 1989), with easier bits being produced earlier than harder bits (Ferreira, 1996; Ferreira & Swets, 2002; Yamashita & Chang, 2001; Stallings, MacDonald, & O’Seaghdha, 1998; see also Lau, & Hwang, 2016), it could be the case that post-modified NPs are easier to produce than pre-modifiers, and therefore more frequently encountered by comprehenders. This might again be due the fact the post-modified NPs conform to the “given-beforenew” principle. Since given information is easier to access from memory than new information, post-modified NPs might be easier to produce than pre-modifiers, which in turn might facilitate their encoding and therefore their subsequent retrieval. It is important to note that, based on some previous findings, the frequency of pre- vs. post-modified NPs could have had the reverse effect on referential processing, with pronouns associated with the less frequent structure forming more salient representations in memory during encoding and resulting in easier subsequent retrieval (see Van Gompel, & Majid, 2004). However, our data clearly did not lend any support to this possibility.
Third, pre-modifying adjectives need to be maintained in memory until the head noun is revealed and before integration can take place. As such, there might be a maintenance cost and/or a delayed integration associated with the processing pre-modified NPs (Wagers & Phillips, 2014), whereas the integration of post-modifying information can start immediately, with no maintenance cost8. These factors might then complicate the encoding of pre-modified NPs compared to post-modified NPs. Future research is needed to distinguish between these three (and potentially other) explanations for the modifier position effect.
As mentioned in the Introduction, one of the potential mechanisms offered by memory-based theories of language processing to explain the semantic richness effect is the head-reactivation account. Based on this account, the memory representation of the head noun becomes reactivated as the modifying information is being encoded (Lewis & Vasishth, 2005; Lewis et al., 2006). Importantly, this explanation has only been shown for cases in which a head category (such as a verb) can be predicted and therefore pre-activated (e.g., Jaeger, Fedorenko, & Gibson, 2005; Konieczny, 2000; Vasishth & Lewis, 2006), which raises the question as to whether prediction is a necessary requirement for head-reactivation. Although our results suggest no evidence that prediction facilitates encoding over and above semantic richness, the head-reactivation account could still explain our results. Specifically, it could be the case that for pre-modifiers, only the syntactic category of the head noun (plus perhaps a little semantic information) is reactivated because this is only information that can be predicted based on the pre-modifying adjectives. However, because the full lexical semantics of the head noun is available in the case of post-modifiers, head-reactivation could involve both syntactic as well as semantic information in the case of post-modified NPs, leading to more robust representations. Note that the head-reactivation account also provides a reasonable explanation for why time-based decay cannot reduce the activation level of post-modified NPs. Time-based decay is an established effect in memory and psycholinguistics based on the fact that memory representations fade (i.e., their activation drops) due to passage of time (Baddeley, 2000; Chomsky, 1965; Gibson, 1998; Lewis & Vasishth, 2005; Lewis et al., 2006; Van Dyke & Lewis, 2003). Interestingly, relative to pre-modified NPs, the head of post-modified NPs was always necessarily farther from the retrieval point, meaning that decay actually worked against post-modified NPs and in favor of pre-modified NPs. As such, the observation that post-modified NPs have higher accessibility than pre-modified NPs requires a mechanism for how the effect of decay is overridden in the case of post-modified NPs. The head-reactivation account provides precisely that mechanism.
Although our results are interpretable in terms of head-reactivation (Lewis & Vasishth, 2005; Lewis et al., 2006), the current memory-based models do not readily provide an explanation for the retrieval ease differences between pre- and post-modified NPs. This is because under the current versions of these models, it is unclear what gets reactivated in the case of modified NPs; is it only syntactic features? Or could both syntactic and semantic features get re-activated, as we suggest in this study? Moreover, the current models are not clear on how exactly predictive processing interacts with reactivation (see Parker, Shvartsman, & Van Dyke, 2017). We believe the results of our study poses interesting challenges for the current memory-based models of language processing, and that resolving these challenges would help improve these models.
One major limitation of our study is that it is entirely based on English where pre-modifications are always limited and bounded to adjectives, and post-modifiers are either relative clauses or prepositional phrases. Some of the possibilities proposed above could be tested by investigating the role of modifier position in languages that allow adjectives to modify a noun both before and after (e.g., Tagalog). This investigation would allow testing how much of the modifier position effect is due to the availability of the head noun. Similarly, studying languages that permit only post-modifiers but through both simple adjectives as well as relative clauses (e.g., Persian) would allow testing the contribution of the type of modifier to the current results. Relatedly, we believe that our results open up new and exciting avenues for investigating the role of memory, information structure, discourse, and perceptual factors in the encoding and retrieval of words during sentence and/or discourse processing. For example, if our results indeed generalize to other languages, how would they contribute to our understanding of encoding and retrieval operations during language processing? Would variations in memory capacity and linguistic knowledge modulate the modifier position effect? We believe that further experimental as well as modelling research investigating these (and similar) questions would considerably improve our understanding of how words and encoded and retrieved during language production and comprehension.
Another limitation of our study is that most of the adjectives that we employed lend themselves to “intersective” as opposed non-intersective interpretations. An intersective interpretation of an adjective applies the quality of the adjective to the real-life referent of the noun, whereas a non-intersective interpretation does not apply the quality to the referent entirely but to some subset of it. For instance, in Arthur was a brave king, the intersective interpretation would be that Arthur is a king and Arthur is brave, whereas a non-intersective interpretation would be that Arthur is brave as a king but not necessarily brave in his other roles (say, as a husband). It is therefore unclear what effect these two particular semantic interpretations might have an impact on the accessibility of associated NPs and future research is required to investigate this issue. A final limitation of our study is that we did not use strongly constraining adjectives, thereby rendering the head nouns not maximally predictable in the case of pre-modified NPs. Future research should look into the potential effect of the constraining power of adjectives on the subsequent accessibility of the head of pre- vs. post-modified NPs.
Conclusion
In four experiments, we demonstrated that post-modified NPs are encoded more robustly in memory than are pre-modifiers, leading to easier subsequent retrieval. These results are the first showing encoding variations as a function of modifier position. Our results contribute to the current memory-based models of language processing by showing that distinctiveness alone cannot account for the retrieval ease of semantically rich NPs, that the effect of time-based decay is overridden by the accessibility conferred by post-modifiers, and also that predictability of the head noun does not facilitate encoding. We suggest that processes related to perception, memory, and information structure are responsible for this effect. Future experimental and modelling research in English and other languages is required to tease apart the relative contributions of these cognitive factors.
Appendix 1. The raw frequencies of talking about NP1, NP2 and excluded responses in Experiment 1.
| Modifier Position | NP Type | choice | Frequency | % NP1 Reference |
|---|---|---|---|---|
| Baseline | Baseline | NP1 | 126 | 33.96 |
| Baseline | Baseline | NP2 | 245 | |
| Baseline | Excluded | Excluded | 109 | - |
| Pre-modified | Target | NP1 | 134 | 37.12 |
| Pre-modified | Target | NP2 | 227 | |
| Pre-modified | Competitor | NP1 | 162 | 39.42 |
| Pre-modified | Competitor | NP2 | 249 | |
| Pre-modified | Excluded | Excluded | 188 | - |
| Post-modified | Target | NP1 | 143 | 41.45 |
| Post-modified | Target | NP2 | 202 | |
| Post-modified | Competitor | NP1 | 167 | 40.73 |
| Post-modified | Competitor | NP2 | 243 | |
| Post-modified | Excluded | Excluded | 205 | - |
Appendix 2. The raw frequencies of pronominal reference to targets and competitors in Experiment 1.
| Modifier Position | NP Type | form | Frequency | % Pronominal Reference |
|---|---|---|---|---|
| Baseline | Baseline | Pronoun | 113 | 30 |
| Baseline | Baseline | Repeated Noun | 258 | |
| Pre-modified | Target | Pronoun | 124 | 34 |
| Pre-modified | Target | Repeated Noun | 237 | |
| Pre-modified | Competitor | Pronoun | 122 | 30 |
| Pre-modified | Competitor | Repeated Noun | 289 | |
| Post-modified | Target | Pronoun | 136 | 39 |
| Post-modified | Target | Repeated Noun | 209 | |
| Post-modified | Competitor | Pronoun | 101 | 25 |
| Post-modified | Competitor | Repeated Noun | 309 |
Appendix 3. The full results for talking about NP1 vs. NP2 in Experiment 1.
| Analysis | Contrast | β | SE | z | p |
|---|---|---|---|---|---|
| Modification Type | PreM_Target vs. Baseline | .09 | .17 | .55 | .58 |
| PostM_Target vs. Baseline | .27 | .17 | 1.58 | .11 | |
| PreM_Target vs. PostM_Target | .18 | .17 | 1.04 | .29 | |
| Relative Richness | Modifier Position | .11 | .10 | 1.12 | .26 |
| NP Type | −.03 | .10 | −.32 | .54 | |
| Modifier Position ×NP Type | .12 | .21 | .60 | .01 | |
| NP Type within PreM | .17 | .17 | .97 | .33 | |
| NP Type within PostM | .04 | .18 | .23 | .81 |
Appendix 4. The raw frequencies of talking about NP1, NP2 and “Other” (i.e., excluded) responses in Experiment 2.
| Modifier Position | NP Type | choice | Frequency | % NP1 Reference |
|---|---|---|---|---|
| Baseline | Baseline | NP1 | 142 | 40.69 |
| Baseline | Baseline | NP2 | 207 | |
| Baseline | other | other | 211 | - |
| Pre-modified | Target | NP1 | 186 | 50.68 |
| Pre-modified | Target | NP2 | 181 | |
| Pre-modified | Competitor | NP1 | 161 | 47.92 |
| Pre-modified | Competitor | NP2 | 175 | |
| Pre-modified | other | other | 417 | - |
| Post-modified | Target | NP1 | 140 | 46.36 |
| Post-modified | Target | NP2 | 162 | |
| Post-modified | Competitor | NP1 | 152 | 41.19 |
| Post-modified | Competitor | NP2 | 217 | |
| Post-modified | other | other | 449 | - |
Appendix 5. The raw frequencies of pronominal reference to NP1 and NP2 in Experiment 2.
| Modifier Position | NP Type | form | Frequency | % Pronominal Reference |
|---|---|---|---|---|
| Baseline | Baseline | Pronoun | 143 | 40.97 |
| Baseline | Baseline | Repeated Noun | 206 | |
| Pre-modified | Target | Pronoun | 191 | 52.04 |
| Pre-modified | Target | Repeated Noun | 176 | |
| Pre-modified | Competitor | Pronoun | 158 | 47.02 |
| Pre-modified | Competitor | Repeated Noun | 178 | |
| Post-modified | Target | Pronoun | 168 | 55.63 |
| Post-modified | Target | Repeated Noun | 134 | |
| Post-modified | Competitor | Pronoun | 158 | 42.82 |
| Post-modified | Competitor | Repeated Noun | 211 |
Appendix 6. The full results for talking about each NP in Experiment 2.
| Analysis | Contrast | β | SE | z | p |
|---|---|---|---|---|---|
| Modification Type | PreM_Target vs. Baseline | .39 | .17 | 2.28 | .03 |
| PostM_Target vs. Baseline | .19 | .18 | 1.05 | .29 | |
| PreM_Target vs. PostM_Target | −.20 | .17 | −1.13 | .25 | |
| Relative Richness | Modifier Position | −.21 | .11 | −1.96 | .49 |
| NP Type | .16 | .11 | 1.45 | .14 | |
| Modifier Position ×NP Type | .09 | .28 | .34 | .73 | |
| NP Type within PreM | −.02 | .18 | −.13 | .89 | |
| NP Type within PostM | −.20 | .25 | −.82 | .41 |
Appendix 7. The raw frequencies of talking about NP1, NP2 and “Other” (i.e., excluded) responses in Experiment 3.
| Modifier Position | NP Type | choice | Frequency | % NP1 Reference |
|---|---|---|---|---|
| Baseline | Baseline | NP1 | 90 | 25.94 |
| Baseline | Baseline | NP2 | 257 | |
| Baseline | other | other | 243 | |
| Pre-modified | Target | NP1 | 63 | 18.81 |
| Pre-modified | Target | NP2 | 272 | |
| Pre-modified | Competitor | NP1 | 77 | 19.64 |
| Pre-modified | Competitor | NP2 | 315 | |
| Pre-modified | other | other | 453 | |
| Post-modified | Target | NP1 | 78 | 22.81 |
| Post-modified | Target | NP2 | 264 | |
| Post-modified | Competitor | NP1 | 73 | 18.72 |
| Post-modified | Competitor | NP2 | 317 | |
| Post-modified | other | other | 448 |
Appendix 8. The raw frequencies of pronominal reference to NP1 and NP2 in Experiment 3.
| Modifier Position | NP Type | form | Frequency | % Pronominal Reference |
|---|---|---|---|---|
| Baseline | Baseline | Pronoun | 216 | 62.25 |
| Baseline | Baseline | Repeated Noun | 131 | |
| Pre-modified | Target | Pronoun | 230 | 68.66 |
| Pre-modified | Target | Repeated Noun | 105 | |
| Pre-modified | Competitor | Pronoun | 256 | 65.31 |
| Pre-modified | Competitor | Repeated Noun | 136 | |
| Post-modified | Target | Pronoun | 240 | 70.18 |
| Post-modified | Target | Repeated Noun | 102 | |
| Post-modified | Competitor | Pronoun | 245 | 62.82 |
| Post-modified | Competitor | Repeated Noun | 145 |
Appendix 9. The full results for talking about each NP in Experiment 3.
| Analysis | Contrast | β | SE | z | p |
|---|---|---|---|---|---|
| Modification Type | PreM_Target vs. Baseline | −.40 | .20 | −2.01 | .04 |
| PostM_Target vs. Baseline | −.17 | .19 | −.92 | .35 | |
| PreM_Target vs. PostM_Target | .22 | .20 | 1.09 | .27 | |
| Relative Richness | Modifier Position | .09 | .13 | .69 | .48 |
| NP Type | .09 | .13 | .74 | .45 | |
| Modifier Position ×NP Type | .30 | .26 | 1.15 | .24 | |
| NP Type within PreM | .13 | .26 | .49 | .62 | |
| NP Type within PostM | −.20 | .19 | −1.05 | .29 |
Appendix 10. Experimental stimuli for Experiments 1 and 2. Only the unmodified (baseline) condition is shown. Pre-modified and Post-modified versions can be made by attaching the modification to the two NPs as preceding adjectives, or as post-modifying relative clauses (using “who was”), respectively. The experimental stimuli for Experiment 2 can be constructed by replacing the underlined NP with the NP to the left of “/”.
| Item | Sentence | Modification |
|---|---|---|
| 1 | The cameraman/actress slapped the actor. | frustrated and visibly upset |
| 2 | The boxer ignored the cowboy/cowgirl. | terribly distressed and anxious |
| 3 | The clergyman/monk supported the monk/godmother. | worried |
| 4 | The choirboy/boy approached the boyscout/girl. | unpopular |
| 5 | The mermaid encountered the goddess/god. | fearless, brave, and adventurous |
| 6 | The stableman bothered the shepherd/shepherdess. | fussy and demanding |
| 7 | The saleswoman contacted the businesswoman/businessman. | rich and successful |
| 8 | The noblewoman blackmailed the countess/count. | struggling |
| 9 | The sorcerer misunderstood the god/goddess. | distracted |
| 10 | The housemaid followed the lady/gentleman. | tall and good-looking |
| 11 | The schoolgirl/schoolboy poked the woman. | lively and energetic |
| 12 | The ballerina/football player photographed the cheerleader. | outgoing and well-connected |
| 13 | The friar supported the congressman/congresswoman. | enthusiastic |
| 14 | The policeman/policewoman accused the detective. | undercover |
| 15 | The emperor/empress betrayed the colonel. | completely weak and powerless |
| 16 | The policewoman/stewardess helped the godmother/godfather. | wise and considerate |
| 17 | The baroness visited the empress/baron. | scared |
| 18 | The deliveryman phoned the milkman/maid. | impatient and terribly irate |
| 19 | The shepherdess/lady congratulated the princess/shepherd. | deeply happy and relieved |
| 20 | The nun attacked the widow/priest. | inexperienced |
| 21 | The governess bribed the mistress/governor. | cunning and appallingly dishonest |
| 22 | The headmaster tripped the governor/headmistress. | lazy and disorganized |
| 23 | The nobleman/heiress misled the lord. | greedy and selfish |
| 24 | The Dutchman/Dutchwoman rescued the sportsman. | strong and muscular |
| 25 | The gunman/ballerina shot the pilot. | heartless |
| 26 | The sorceress/wizard killed the witch. | cold and ruthless |
| 27 | The baron/heroine defied the bishop. | panicked and incompetent |
| 28 | The bachelorette spotted the girl/bachelor. | smart and vigilant |
| 29 | The chairwoman/chairman argued with the duchess. | obstinate and confrontative |
| 30 | The camerawoman/cameraman welcomed the actress/camerawoman. | world-famous |
| 31 | The wizard/sorceress trapped the sailor. | loyal and patriotic |
| 32 | The choirgirl/choirboy photographed the bridesmaid. | creative and artistic |
| 33 | The housewife mistreated the maid/milkman. | mentally ill |
| 34 | The barmaid missed the waitress/waiter. | kind and caring |
| 35 | The clergyman supported the monk/priestess. | knowledgeable and dedicated |
| 36 | The salesgirl/salesman consulted with the nanny. | sad and lonely |
| 37 | The horseman intimidated the huntsman/queen. | courageous |
| 38 | The fisherman negotiated with the gentleman/salesgirl. | highly cautious |
| 39 | The anchorman debated with the spokesman/spokeswoman. | eloquent and articulate |
| 40 | The prince/princess criticized the king. | cruel and merciless |
Appendix 11. Experimental stimuli for Experiment 3.
| Item | Sentence | NP1 modification | NP2 modification |
|---|---|---|---|
| 1 | The carpet mesmerized the photographer for hours in the museum. | ancient and ornate | stylish and artistic |
| 2 | The castle pleased the knight after the long and dangerous expedition. | clean and orderly | prim and proper |
| 3 | The necklace had sentimental value for the miner. | fine and golden | lonely and emotional |
| 4 | The chain disgusted the stuntman. | slimy and filthy | pampered |
| 5 | The mural delighted the painter during the exhibition. | lively and colorful | internationally famous |
| 6 | The fire strengthened the hunter on the cold winter night. | warm and crackling | scared |
| 7 | The statue resembled the warrior from a distance. | marble and detailed | brave and crafty |
| 8 | The wave made the surfer very anxious. | enormous and rapid | naive and reckless |
| 9 | The helicopter reached the tourist within the hour. | swift and agile | distraught and injured |
| 10 | The rope frustrated the climber during the descent. | tangled and slippery | skilled and professional |
| 11 | The poll infuriated the politician greatly. | crucial and determining | rich and powerful |
| 12 | The tank startled the protestor a couple of times. | noisy and rumbling | meek and disoriented |
| 13 | The tractor terrified the cyclist. | rusty and sharp-bladed | drunk |
| 14 | The plan baffled the councilor quite a bit. | extensive and highly complicated | liberal and environmental |
| 15 | The battleship frightened the sniper all of a sudden. | huge and heavily armed | unsuspecting and overconfident |
| 16 | The raft rescued the mountaineer in the raging river. | strong and well-made | old and bearded |
| 17 | The newspaper undermined the fraudster at long last. | small but independent | dastardly |
| 18 | The ambulance took the boy to the hospital in time. | reliable and well-equipped | unconscious |
| 19 | The nail enabled the prisoner to break free. | long and narrow | patient and persistent |
| 20 | The noise made the guard suspicious in the dark night. | buzzing and strange | tired and cranky |
| 21 | The dungeon intimidated the slave at first sight. | dark and daunting | helpless and abused |
| 22 | The motorbike followed the lady for the whole day. | stealthy and quiet | short and brunette |
| 23 | The jeep transported the soldier across the battle field. | big and heavy-duty | wounded and bleeding |
| 24 | The cabin calmed the peasant during the storm. | sturdy, well-built and cozy | frail and vulnerable |
| 25 | The decision influenced the resident quite clearly. | hasty and irrational | newly-settled |
| 26 | The wagon irritated the cowboy all night long. | loud and extremely squeaky | exhausted |
| 27 | The cage guarded the girl during the raid. | empty and spacious | terribly frightened |
| 28 | The salary offended the executive a great deal. | unexpectedly low | cocky and entitled |
| 29 | The letter disillusioned the queen. | confidential and informative | kind and sympathetic |
| 30 | The gift made the youngster happy at the party. | expensive and unexpected | shy and reserved |
| 31 | The legislation disappointed the immigrant a great deal. | unfair and prodiscrimination | hopeful and hardworking |
| 32 | The gate nudged the doorman a bit. | wide open and wooden | distracted and confused |
| 33 | The plane endangered the pilot quite seriously. | new and untested | newly-trained |
| 34 | The island astonished the mermaid at the end of the long journey. | stunningly beautiful | miserable and desperate |
| 35 | The portrait amused the artist quite noticeably. | priceless and masterfully drawn | very curious |
| 36 | The condition satisfied the applicant in most cases. | realistic and sensible | unbelievably ambitious |
| 37 | The stained glass enchanted the villager for hours on end. | intricate and vibrant | humble and pious |
| 38 | The mannequin shocked the passer-by a couple of times. | eerily and lifelike | mentally unstable |
| 39 | The salad impressed the customer yesterday. | fresh and organic | adventurous and gluttonous |
| 40 | The moon inspired the novelist. | bright and full | young and promising |
| 41 | The report confused the officer throughout the investigation. | poorly written and unconfirmed | belligerent and short-tempered |
| 42 | The boat stressed out the swimmer in the lagoon. | seemingly out-of-control and swerving | poor and timid |
| 43 | The song entertained the teenager throughout the night. | catchy and widely popular | homeless and disheveled |
| 44 | The food poisoned the diner quite seriously. | weird and sour-tasting | slim and petite |
| 45 | The sandwich worried the grandma during the lunch rush. | sloppily-made and discolored | finicky and picky |
| 46 | The cookie made the child salivate profusely. | aromatic and delicious-looking | hungry and hyper-active |
| 47 | The magazine disturbed the newsreader yesterday. | alien and occult | unbelievably paranoid |
| 48 | The steamboat troubled the sailor for the whole day. | large, heavy and slow | crazy and overweight |
| 49 | The vehicle provoked the driver into an accident. | damaged and speeding | rushed and frantic |
| 50 | The sled carried the Eskimo during the hunting trip. | traditional but very efficient | unrelenting and muscular |
Appendix 12. Experimental stimuli for Experiment 4.
| Item | Sentence 1/Sentence 2 | Modification/Question |
|---|---|---|
| 1 | The actor received an Oscar award last year. | very talented and ambitious |
| 1 | In a funny accident, however, he tripped on the red carpet. | He received an Oscar award a couple of years ago. |
| 2 | The cowgirl counted the tumbleweeds that rolled on by. | bored and listless |
| 2 | After a while, she gradually started to feel thirsty. | She started to feel sleepy. |
| 3 | The monk was in a cleaning frenzy. | severely anxious and frantic |
| 3 | By the end of the day, he made the monastery shine like a beacon. | No Question |
| 4 | The mermaid followed the current all the way to the magic island. | adventurous and daring |
| 4 | Right before sunrise, she could finally see the palm trees moving in the wind. | She wanted to go to an island. |
| 5 | The salesgirl found a new job. | extremely rude and distracted |
| 5 | But after only six weeks, she was fired and had to start over. | She lost her job very soon. |
| 6 | The businessman was going bankrupt. | impulsive and emotional |
| 6 | To everyone’s surprise, he kept blaming other people for the predicament. | He refused to take responsibility for the bankruptcy. |
| 7 | The housemaid had the face of an angel. | kind and innocent |
| 7 | In a lucky incident, she was invited to model for a fashion magazine. | She was invited to act in a romance movie. |
| 8 | The schoolboy could never finish a test. | slow and annoying |
| 8 | After a few minutes into a test, he would start to dream of adventure and mischief. | He was very focused. |
| 9 | The actress wanted more money for the movie. | greedy and selfish |
| 9 | However, after a long discussion, she finally decided to sign the contract. | No Question |
| 10 | The boxer did not want to fight. | scared and unprepared |
| 10 | As an excuse, he told the press his arm was injured. | He told the truth to the press. |
| 11 | The godmother donated a lot of money to the hospital. | caring and selfless |
| 11 | Many years later, she had a successful heart surgery in the same hospital. | No Question |
| 12 | The boy was exceptionally good at math. | horribly mean and evil |
| 12 | After some research, it turned out that he had a psychological condition. | He had an astonishing talent in math |
| 13 | The girl went shopping on Sunday morning. | rich and fashionable |
| 13 | When the sun was setting, she finally finished shopping and returned home. | No Question |
| 14 | The king was diagnosed with a terminal disease. | cruel and merciless |
| 14 | For almost two months, he had no idea because no one would dare break the news to him. | No Question |
| 15 | The kitchenmaid made an ethnic dinner. | calm and organized |
| 15 | From the color of it, though, she realized the most important ingredient was missing. | No Question |
| 16 | The nun took care of the orphanage | trustworthy and motherly |
| 16 | According to the kids, she was the nicest person on Earth. | She was in charge of a day care center. |
| 17 | The football player was feeling a little dizzy. | overwhelmed and nervous |
| 17 | At the end of the game, he collapsed on the bench while walking. | No Question |
| 18 | The cheerleader had found a new hobby. | creative and artistic |
| 18 | Every day after work, she would collect leaves and petals. | No Question |
| 19 | The congressman made many bold promises. | power-hungry and misleading |
| 19 | After the elections were over, he was not the same person at all. | No Question |
| 20 | The friar managed the huge ceremony. | smooth and savvy |
| 20 | For the whole following week, he was the talk of the town. | He supervised a big ceremony. |
| 21 | The policeman knocked on the door. | polite and civilized |
| 21 | With the strange noise, he was sure that something was going on in the house. | No Question |
| 22 | The detective was on a very difficult case. | shrewd and extremely intelligent |
| 22 | In an astonishing record, he solved it in just 24 hours. | No Question |
| 23 | The empress ruled for 50 years. | prosperous and affluent |
| 23 | When the kingdom fell, she committed suicide in prison. | No Question |
| 24 | The colonel led the war to victory. | strict and authoritative |
| 24 | After the war was formally over, he received a medal of honor from the president. | No Question |
| 25 | The stewardess had never seen such a strong wind during a flight. | poised and experienced |
| 25 | When the wings started fluttering, she started to feel very terrified. | No Question |
| 26 | The godfather was taking a trip across the country. | impatient and ill-tempered |
| 26 | Even at the destination, he never stopped complaining. | No Question |
| 27 | The baron was getting married. | humble and modest |
| 27 | When the ring went missing on the wedding day, he started to panic and broke his arm while searching for it. | No Question |
| 28 | The lady strolled through the park. | carefree and relaxed |
| 28 | On bright and sunny days, she would always take a walk in the local park. | No Question |
| 29 | The shepherd rounded up the flock. | careful and thorough |
| 29 | With the dense and dark clouds, he wanted to make sure the sheep will be safe. | No Question |
| 30 | The priest left home in the early morning. | pious and devout |
| 30 | For the Sunday service, he would always get very excited. | No Question |
| 31 | The woman refused to listen. | biased and uncaring |
| 31 | According to everyone, she was totally beyond reasoning. | No Question |
| 32 | The ballerina performed during a dance festival. | graceful and enchanting |
| 32 | After the show, she received an standing ovation from the audience. | She performed during Christmas celebrations. |
| 33 | The widow was very unhappy. | sorrowful and miserable |
| 33 | For 20 years, she had been unsuccessful in finding the right man. | No Question |
Footnotes
In this study, we use the terms “accessibility” and “retrieval ease” interchangeably, and both terms refer to how easily a previously encoded representation could be re-accessed or retrieved from memory.
Under some versions of the memory-based retrieval theories, a reactivation is considered a retrieval (see Anderson, Bothell, Byrne, Douglass, Lebiere, & Qin, 2004).
It might be argued that a fourth reason why post-modifiers might facilitate subsequent retrieval is that post-modifying relative clauses result in more syntactically complex structures relative to pre-modifying adjectives (at least in English), and might therefore simply elicit greater processing effort at encoding. Although this is a logical possibility, we do not discuss it further in the present paper because previous research has already shown that processing effort at encoding does not determine ease of subsequent retrieval (Hofmeiser, 2011)
Note that lack of reference to either of the NPs might imply extremely low accessibility for both NPs, and reference to both NPs might imply high and virtually equal accessibility for both NPs. However, since the participants’ task was to simply provide a continuation for the given sentences, such responses probably also reflect an attempt to maximize coherence between the prompt sentence and the response for a specific participant, for a specific item and at a specific time. We therefore believe it is safe to exclude such responses. Moreover, excluding such responses is fully consistent with previous research employing the sentence continuation paradigm (Arnold, 2001; Fukumura & Van Gompel, 2010; Stevenson et al., 1994).
We analyzed the rate of excluded responses as a function of Modifier Position. The results showed no reliable difference between the Post-modified and the Baseline conditions (p = .36), or between the Pre- and Post-modified conditions (p = .18). However, there were significantly more excluded responses in the Pre-modified than the Baseline condition (p = .04). Since we did not observe this effect between the conditions in our next experiments (see below), we suspect this finding is spurious.
More data were lost in this experiment compared to Experiment 1 because, in many trials, the participants pressed the spacebar before or during speaking their responses, resulting in additional loss of data. As in Experiment 1, we also analyzed the rate of excluded responses across conditions, but we did not observe any reliable differences (Pre-modified vs. Baseline: p = .98, Post-modified vs. Baseline: p = .44, Pre-modified vs. Post-modified: p =.27).
As in Experiment 2, we lost additional data in this experiment due to participants terminating the recordings too early (Experiments 2 and 3 were run simultaneously, but the participants were not the same). There were no reliable difference in the probability of excluded responses across the conditions. (Pre-modified vs. Baseline: p = .26, Post-modified vs. Baseline: p = .29, Pre-modified vs. Post-modified: p =.88)
Note that this maintenance cost and/or delayed integration does not seem to be strong enough to reverse the semantic richness effect. This is because our first three experiments as well as previous researchers have demonstrated easier retrieval for pre-modified than for unmodified NPs (Hofmeister, 2011).
References
- Almor A (1999). Noun-phrase anaphora and focus: The informational load hypothesis. Psychological Review, 106, 748–765. [DOI] [PubMed] [Google Scholar]
- Anderson JR, Bothell D, Byrne MD, Douglass S, Lebiere C, & Qin Y (2004). An integrated theory of the mind. Psychology Review, 111, 1036–1060. [DOI] [PubMed] [Google Scholar]
- Ariel M (1990). Accessing noun-phrase antecedents. London: Routledge. [Google Scholar]
- Ariel M (1991). The function of accessibility in a theory of grammar. Journal of Pragmatics, 16, 443–463. [Google Scholar]
- Ariel M (1996). Referring expressions and the ± coreference distinction In Fretheim JKGT (Ed.), Reference and referent accessibility (pp. 13–35). Amsterdam, The Netherlands: Benjamins. [Google Scholar]
- Arnett N, & Wagers M (2017). Subject encodings and retrieval interference. Journal of Memory and Language, 93, 22–54. [Google Scholar]
- Arnold JE (2001). The effect of thematic roles on pronoun use and frequency of reference continuation. Discourse Processes, 31, 137–162. [Google Scholar]
- Arnold JE, & Griffin Z (2007). The effect of additional characters on choice of referring expression: Everyone counts. Journal of Memory and Language, 56, 521–536. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Baddeley AD (1986). Working memory. Oxford, UK: Clarendon Press. [Google Scholar]
- Baddeley A (2000). The episodic buffer: a new component of working memory?. Trends in cognitive sciences, 4(11), 417–423. [DOI] [PubMed] [Google Scholar]
- Baker L, & Wagner JL (1987). Evaluating information for truthfulness: the effects of logical subordination. Memory & Cognition, 15, 247e255. [DOI] [PubMed] [Google Scholar]
- Barr DJ, Levy R, Scheepers C, & Tily HJ (2013). Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of memory and language, 68(3), 255–278. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Benatar A, & Clifton C Jr (2014). Newness, givenness and discourse updating: Evidence from eye movements. Journal of Memory and Language, 71(1), 1–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bock JK (1982). Toward a cognitive psychology of syntax: information processing contributions to sentence formulation. Psychological Review, 89, 1–47. [Google Scholar]
- Bock JK, & Warren RK (1985). Conceptual accessibility and syntactic structure in sentence formulation. Cognition, 21(1), 47–67. [DOI] [PubMed] [Google Scholar]
- Branigan HP, Pickering MJ, & Tanaka M (2008). Contributions of animacy to grammatical function assignment and word order during production. Lingua, 118(2), 172–189. [Google Scholar]
- Brennan SE (1995). Centering attention in discourse. Language and Cognitive Processes, 10, 137–167. [Google Scholar]
- Brennan SE, Friedman MW, & Pollard CJ (1987, July). A centering approach to pronouns In Proceedings of the 25th annual meeting on Association for Computational Linguistics(pp. 155162). Association for Computational Linguistics. [Google Scholar]
- Chafe WL (1976). Givenness, contrastiveness, definiteness, subjects, topics, and point of view In Li CN (Ed.), Subject and topic (pp. 25–56). New York, NY: Academic Press. [Google Scholar]
- Chang F, Dell GS, & Bock K (2006). Becoming syntactic. Psychological review, 113(2), 234. [DOI] [PubMed] [Google Scholar]
- Chang F, Dell GS, Bock K, & Griffin ZM (2000). Structural priming as implicit learning: A comparison of models of sentence production. Journal of Psycholinguistic Research, 29(2), 217–230. [DOI] [PubMed] [Google Scholar]
- Chomsky N (1965). Aspects of the Theory of Syntax. MIT Press [Google Scholar]
- Chomsky N (1986). Knowledge of language: Its nature, origin, and use. Greenwood Publishing Group. [Google Scholar]
- Chomsky N (1993). Lectures on government and binding: The Pisa lectures (No. 9). Walter de Gruyter. [Google Scholar]
- Chomsky N (1995). Categories and transformations. The minimalist program, 219, 394. [Google Scholar]
- Christianson K, Hollingworth A, Halliwell JF, & Ferreira F (2001). Thematic roles assigned along the garden path linger. Cognitive Psychology, 42, 368–407. [DOI] [PubMed] [Google Scholar]
- Christianson K, Williams CC, Zacks RT, & Ferreira F (2006). Misinterpretations of garden-path sentences by older and younger adults. Discourse Processes, 42, 205–238. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Clark A (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3), 181–204. [DOI] [PubMed] [Google Scholar]
- Clark E, & Clark H (1978). Universals, relativity and language processing In Greenberg J (Ed.), Universals of human language (Method and Theory, Vol. 1, pp. 225–277). Stanford: Stanford University Press. [Google Scholar]
- Conway CM, Bauernschmidt A, Huang SS, & Pisoni DB (2010). Implicit statistical learning in language processing: Word predictability is the key. Cognition, 114(3), 356–371. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cook AE (2014). Processing anomalous anaphors. Memory & Cognition, 42(7), 1171–1185. [DOI] [PubMed] [Google Scholar]
- De Smedt K, & Kempen G (1987). Incremental sentence production, self-correction and coordination In Natural language generation (pp. 365–376). Springer, Dordrecht. [Google Scholar]
- Dell GS, McKoon G, & Ratcliff R (1983). The activation of antecedent information during the processing of anaphoric reference in reading. Journal of Verbal Learning and Verbal Behavior, 22, 121–132. [Google Scholar]
- Dillon BW (2011). Structured access in sentence comprehension. PhD dissertation, University of Maryland. [Google Scholar]
- Dillon B, Chow WY, Wagers M, Guo T, Liu F, & Phillips C (2014). The structure-sensitivity of memory access: evidence from Mandarin Chinese. Frontiers in psychology, 5, 1025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Dillon B, Clifton C Jr., & Frazier L (2014). Pushed aside: Parentheticals, memory and processing. Language, Cognition and Neuroscience, 29, 483–498. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Dillon B, Clifton C Jr, Sloggett S, & Frazier L (2017). Appositives and their aftermath: Interference depends on at-issue vs. not-at-issue status. Journal of Memory and Language, 96, 93–109. [Google Scholar]
- Ferreira VS (1996). Is it better to give than to donate? Syntactic flexibility in language production. Journal of memory and language, 35(5), 724–755. [Google Scholar]
- Ferreira F (2003). The misinterpretation of noncanonical sentences. Cognitive psychology, 47(2), 164–203. [DOI] [PubMed] [Google Scholar]
- Ferreira F, Bailey KG, & Ferraro V (2002). Good-enough representations in language comprehension. Current Directions in Psychological Science, 11, 11–15. [Google Scholar]
- Ferreira F, Christianson K, & Hollingworth A (2001). Misinterpretations of garden-path sentences: implications for models of sentence processing and reanalysis. Journal of Psycholinguistic Research, 30, 3–20. [DOI] [PubMed] [Google Scholar]
- Ferreira F, & Henderson JM (1990). Use of verb information in syntactic parsing: Evidence from eye movements and word-by-word self-paced reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16(4), 555–568. doi: 10.1037/0278-7393.16.4.555 [DOI] [PubMed] [Google Scholar]
- Ferreira F, & Lowder MW (2016). Prediction, information structure, and good-enough language processing In Psychology of Learning and Motivation (Vol. 65, pp. 217–247). Academic Press. [Google Scholar]
- Ferreira F, & Swets B (2002). How incremental is language production? Evidence from the production of utterances requiring the computation of arithmetic sums. Journal of Memory and Language, 46(1), 57–84. [Google Scholar]
- Fletcher CR (1984). Markedness and topic continuity in discourse processing. Journal of Verbal Learning and Verbal Behavior, 23, 487–493. [Google Scholar]
- Foraker S, & McElree B (2007). The role of prominence in pronoun resolution: Active versus passive representations. Journal of Memory and Language, 56(3), 357–383. [Google Scholar]
- Fukumura K, & Van Gompel RPG (2010). Choosing anaphoric expressions: Do people take into account likelihood of reference? Journal of Memory and Language, 62, 52–66. [Google Scholar]
- Fukumura K, & Van Gompel RPG (2011). The effects of animacy in the choice of referring expressions. Language and Cognitive Processes, 26, 1472–1504. [Google Scholar]
- Gallo DA, Meadow NG, Johnson EL, & Foster KT (2008). Deep levels of processing elicit a distinctiveness heuristic: Evidence from the criterial recollection task. Journal of Memory and Language, 58(4), 1095–1111. [Google Scholar]
- Gelman A, & Hill J (2006). Data analysis using regression and multilevel/hierarchical models. Cambridge University Press. [Google Scholar]
- Gernsbacher MA (1989). Mechanisms that improve referential access. Cognition, 32, 99–156. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Gernsbacher MA, & Hargreaves DI (1988). Accessing sentence participants: The advantage of first mention. Journal of Memory and Language, 27, 699–711. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Gerrig RJ, & McKoon G (1998). The readiness is all: The functionality of memory-based text processing. Discourse Processes, 26, 67–86. [Google Scholar]
- Gibson EA (1998) Linguistic complexity: locality of syntactic dependencies. Cognition, 68, 1–76 [DOI] [PubMed] [Google Scholar]
- Givón T (1983). Topic continuity in discourse: An introduction In Givón T (Ed.), Topic continuity in discourse: A quantitative cross-language study (pp. 1–41). Amsterdam: John Benjamins. [Google Scholar]
- Gordon PC, Grosz BJ, & Gilliom LA (1993). Pronouns, names, and the centering of attention in discourse. Cognitive Science, 17, 311–347. [Google Scholar]
- Grosz BJ, Joshi AK, & Weinstein S (1995). Centering: A framework for modelling the local coherence of discourse. Computational Linguistics, 21, 203–225. [Google Scholar]
- Gundel JK, Hedberg N, & Zacharski R (1993). Cognitive status and the form of anaphoric expressions in discourse. Language, 69, 274–307. [Google Scholar]
- Hale J (2001, June). A probabilistic Earley parser as a psycholinguistic model In Proceedings of the second meeting of the North American Chapter of the Association for Computational Linguistics on Language technologies (pp. 1–8). Association for Computational Linguistics. [Google Scholar]
- Hale J (2006). Uncertainty about the rest of the sentence. Cognitive Science, 30(4), 643–672. [DOI] [PubMed] [Google Scholar]
- Halliday M (1967). Notes on transitivity and theme in English, Part II. Journal of Linguistics, 3, 199–244. [Google Scholar]
- Haviland SE, & Clark HH (1974). What’s new? Acquiring new information as a process in comprehension. Journal of verbal learning and verbal behavior, 13(5), 512–521. [Google Scholar]
- Hofmeister P (2011). Representational complexity and memory retrieval in language comprehension. Language and Cognitive Processes, 26, 376–405. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hofmeister P, & Vasishth S (2014). Distinctiveness and encoding effects in online sentence comprehension. Frontiers in Psychology, 5, 1237 10.3389/fpsyg.2014.01237. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hopman EW, & MacDonald MC (2018). Production practice during language learning improves comprehension. Psychological science, 0956797618754486. [DOI] [PubMed] [Google Scholar]
- Jäger LA, Engelmann F, & Vasishth S (2017). Similarity-based interference in sentence comprehension: Literature review and Bayesian meta-analysis. Journal of Memory and Language, 94, 316–339. [Google Scholar]
- Jacoby L, & Craik F (1979). “Effectsofelaborationofprocessingatencoding and retrieval: trace distinctiveness and recovery of initial context,”in Levels of Processing in Human Memory, eds Cermak L and Craik F (Hillsdale, NJ: Lawrence Erlbaum Associates; ),1–21. [Google Scholar]
- Jaeger TF (2010). Redundancy and reduction: Speakers manage syntactic information density. Cognitive Psychology, 61, 23–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jaeger F, Fedorenko E, & Gibson E (2005). Dissociation between production and comprehension complexity In Poster presentation at the 18th CUNY sentence processing conference, University of Arizona [Google Scholar]
- Just MA, & Carpenter PA (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99(1), 122–149. [DOI] [PubMed] [Google Scholar]
- Karimi H, & Ferreira F (2016a). Informativity renders a referent more accessible: Evidence from eyetracking. Psychonomic Bulletin & Review, 23, 507–525. [DOI] [PubMed] [Google Scholar]
- Karimi H, & Ferreira F (2016b). Good-enough linguistic representations and online cognitive equilibrium in language processing. The Quarterly Journal of Experimental Psychology, 69, 1013–1040. [DOI] [PubMed] [Google Scholar]
- Karimi H, Fukumura K, Ferreira F, & Pickering MJ (2014). The effect of noun phrase length on the form of referring expressions. Memory & Cognition, 42, 993–1009. [DOI] [PubMed] [Google Scholar]
- Karimi H, Swaab TY, & Ferreira F (2018). Electrophysiological evidence for an independent effect of memory retrieval on referential processing. Journal of Memory and Language, 102, 68–82. [Google Scholar]
- Kempen G, & Hoenkamp E (1987). An incremental procedural grammar for sentence formulation. Cognitive science, 11(2), 201–258. [Google Scholar]
- Kleinschmidt DF, Fine AB, & Jaeger TF (2012, January). A belief-updating model of adaptation and cue combination in syntactic comprehension. In Proceedings of the Annual Meeting of the Cognitive Science Society (Vol. 34, No. 34). [Google Scholar]
- Konieczny L (2000). Locality and parsing complexity. Journal of Psycholinguistic Research, 29(6), 627–645. [DOI] [PubMed] [Google Scholar]
- Kush D (2013). Respecting relations: Memory access and antecedent retrieval in incremental sentence processing. PhD dissertation, University of Maryland. [Google Scholar]
- Lau SH, & Hwang H (2016). The effects of frequency on pronoun production. Journal of Cognitive Science, 17(4), 547–569. [Google Scholar]
- Levelt WJM (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press. [Google Scholar]
- Levy R (2008). Expectation-based syntactic comprehension. Cognition, 106(3), 1126–1177. [DOI] [PubMed] [Google Scholar]
- Lewis RL, & Vasishth S (2005). An activation-based model of sentence processing as skilled memory retrieval. Cognitive Science, 29, 375–419. [DOI] [PubMed] [Google Scholar]
- Lewis RL, Vasishth S, & Van Dyke JA (2006). Computational principles of working memory in sentence comprehension. Trends in Cognitive Sciences, 10, 447–454. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lucas MM, Tanenhaus MK, & Carlson GN (1990). Levels of representation in the interpretation of anaphoric reference and instrument inference. Memory & Cognition, 18, 611–631. [DOI] [PubMed] [Google Scholar]
- MacDonald MC (2013). How language production shapes language form and comprehension. Frontiers in psychology, 4, 226. [DOI] [PMC free article] [PubMed] [Google Scholar]
- MacDonald MC (2016). Speak, act, remember: The language-production basis of serial order and maintenance in verbal memory. Current Directions in Psychological Science, 25(1), 47–53. [Google Scholar]
- MacDonald MC, & MacWhinney B (1990). Measuring inhibition and facilitation from pronouns. Journal of Memory and Language, 29, 469–492. [Google Scholar]
- McDonald SA, & Shillcock RC (2003). Low-level predictive inference in reading: The influence of transitional probabilities on eye movements. Vision Research, 43(16), 1735–1751. [DOI] [PubMed] [Google Scholar]
- Nairne JS (2006). Modeling distinctiveness: Implications for general memory theory. Distinctiveness and memory, 27–46. [Google Scholar]
- Parker D, Shvartsman M, & Van Dyke JA (2017). The cue-based retrieval theory of sentence comprehension: New findings and new challenges Language processing and disorders. Newcastle: Cambridge Scholars Publishing. [Google Scholar]
- Rosenbach A (2008). Animacy and grammatical variation—Findings from English genitive variation. Lingua, 118(2), 151–171. [Google Scholar]
- Sanford AJ, & Garrod SC (1981). Understanding written language. Chichester, UK: Wiley. [Google Scholar]
- Silverstein M (1976). Hierarchy of features an ergativity In Dixon RMW (Ed.), Grammatical categories in Australian languages (pp. 112–171). Canberra, Australia: Humanities Press. [Google Scholar]
- Sanford AJ, & Garrod S (1989). What, when and how: Questions of immediacy in anaphoric reference resolution. Language and Cognitive Processes, 4, 235–263. [Google Scholar]
- Sanford AJ, & Garrod SC (2005). Memory-based approaches and beyond. Discourse Processes, 39, 205–224. [Google Scholar]
- Sanford AJ, & Sturt P (2002). Depth of processing in language comprehension: Not noticing the evidence. Trends in Cognitive Sciences, 6, 382–386. [DOI] [PubMed] [Google Scholar]
- Stallings LM, MacDonald MC, & O’Seaghdha PG (1998). Phrasal ordering constraints in sentence production: Phrase length and verb disposition in heavy-NP shift. Journal of Memory and Language, 39(3), 392–417. [Google Scholar]
- Stevenson RJ, Crawley RA, & Kleinman D (1994). Thematic roles, focus and the representation of events. Language and Cognitive Processes, 9, 519–548. [Google Scholar]
- Troyer M, Hofmeister P, & Kutas M (2016). Elaboration over a discourse facilitates retrieval in sentence processing. Frontiers in Psychology, 7 10.3389/fpsyg.2016.00374. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Trueswell JC, Tanenhaus MK, & Kello C (1993). Verb-specific constraints in sentence processing: Separating effects of lexical preference from garden-paths. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19(3), 528–553. doi: 10.1037/0278-7393.19.3.528 [DOI] [PubMed] [Google Scholar]
- Van Dyke JA, & Lewis RL (2003). Distinguishing effects of structure and decay on attachment and repair: A cue-based parsing account of recovery from misanalyzed ambiguities. Journal of Memory and Language, 49, 285–316. [Google Scholar]
- Van Dyke JA, & McElree B (2011). Cue-dependent interference in comprehension. Journal of Memory and Language, 65, 247–263. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Van Gompel RP, & Majid A (2004). Antecedent frequency effects during the processing of pronouns. Cognition, 90(3), 255–264. [DOI] [PubMed] [Google Scholar]
- Vasishth S, & Lewis RL (2006). Argument-head distance and processing complexity: Explaining both locality and anti-locality effects. Language, 82(4), 767–794. [Google Scholar]
- Wagers MW, & Phillips C (2014). Going the distance: memory and control processes in active dependency construction. The Quarterly Journal of Experimental Psychology, 67(7), 1274–1304. [DOI] [PubMed] [Google Scholar]
- Yamashita H, & Chang F (2001). “Long before short” preference in the production of a head-final language. Cognition, 81(2), B45–B55. [DOI] [PubMed] [Google Scholar]