Edited By Susanne Fuchs, Daniel Pape, Caterina Petrone and Pascal Perrier
Psycholinguistics and Planning: A Focus on Individual Differences (Benjamin Swets)
Abstract: Researchers in the field of psycholinguistics, and especially language production, tend to use experimental research methods to test theories of models of processing. In doing so, we sometimes overlook systematic variance in task performance that is due to individual differences. One area that could benefit from more work on individual differences is in research concerning the mental mechanisms governing the scope of advance speech planning. In this chapter, I will summarize some of the research I have conducted with colleagues that has explored the utility of the individual differences approach. First, I will show how individual differences approaches can capture a good deal of variance that other more traditional variables might miss. I will then offer some data consistent with the idea that the scope of planning in language production varies not just across experimentally manipulated conditions, but also among individuals. Following this I will argue that this individual differences approach allows for some theoretical advances regarding the general role of working memory in language processing. I will conclude by outlining additional opportunities to conduct individual differences research in language production, with some notes to take appropriate caution doing so.
1. Psychology and variance
The overall purpose of cognitive psychological research is to discover systematic variance in behaviors that can help us to infer the nature of our mental processes and representations. In psycholinguistics, researchers have typically proceeded by searching for systematic variance across situations by manipulating independent variables experimentally. The virtues of this approach are clear, in that they provide the means to assess the causal relationships between variables that allow researchers to arrive at meaningful explanations (models and theories) of psycholinguistic phenomena. The focus of this chapter is on systematic variance in language processing among individuals (individual differences). ← 89 | 90 →
The well known limitation of the individual differences approach is that it is inherently correlational. By merely associating variables rather than manipulating them, researchers fall short of causal explanations. But this type of research also does something very well that the typical psycholinguistic experiment cannot: It can systematically account for variance that occurs among individuals, which generally is error variance (variance that cannot be accounted for) in most experiments. To illustrate the utility of the individual differences approach in psycholinguistics, I presently review a case in which individual differences approaches yielded insights that experimental approaches alone could not have provided. We will begin in the domain of sentence comprehension and then turn to sentence production.
2. An illustrative example
One theoretical focus of psycholinguistic research on sentence comprehension has been on the manner in which a parser decides what to do with new, ambiguous constituents. Frazier (1987), as part of Garden Path Theory, a modular, syntax-first account of parsing, postulated Late Closure, a universal parsing principle: “If grammatically possible, attach new items into the clause or phrase currently being processed.” Take sentence (1) below. One possible parse of this sentence is to associate the relative clause with the first noun phrase (NP1), “the sister”. This “high” attachment interpretation implies that it was the sister who shot herself on the balcony. According to research by Frazier (1979), English speakers instead prefer to associate the relative clause with the second noun phrase (NP2). According to this “low” type of attachment, it was the actress who shot herself on the balcony. According to Frazier’s Garden Path model, this preference exists because the Late Closure heuristic makes a decision based on syntax alone that the relative clause must be part of the currently “open” phrase, which in the case of (1) is “the actress”. Importantly, Garden Path model posited that such parsing strategies should be universal, holding that all languages ought to show the same preference in similar constructions.
Cuetos and Mitchell (1988) showed that other languages, such as Spanish and Dutch, showed an NP1 preference (Cuetos and Mitchell, 1988; Brybaert and Mitchell, 1996). Because preferences varied across languages, this line of research undermined that assumption of Late Closure as a universal parsing strategy. Without such universality, the viability of the syntax-first Garden Path model was reduced.
Subsequent research has demonstrated that despite the theoretical importance of these cross-linguistic differences, there is even more variability in attachment preferences among individual speakers of the same language (Swets et al., 2007). In the research that showed these results, we conducted two studies. In the first, we administered a reading span task to measure working memory and an offline relative clause attachment task to each participant from English speaking (n = 150) and Dutch speaking (n = 96) populations. The tasks were administered to large samples of subjects because the statistical analyses we were conducting, including factor analysis and structural equation modeling, required large samples to detect theoretically interesting effects. For the relative clause attachment task, participants viewed sentences such as (1) on a screen, and then were asked forced choice questions about them that indicated a NP1 or NP2 attachment decision.
2) Who was shot on the balcony? (the sister / the actress)
The reading span task we administered was a modified version of the Dane-man and Carpenter (1980) task. In the task, participants tried to remember lists of 3 to 6 words as they judged whether a series of sentences made sense. Underneath each sentence that they read on the screen in front of them, a word appeared in red. Participants were to circle YES on an answer sheet if the sentence made sense, and NO if it did not, and remember the word in red for later. After 3 to 6 sentence judgments, three question marks appeared on the screen, and participants were to turn the page on the answer sheet and write each of the red words down in the order in which they appeared.
We found that reading span predicted attachment preferences, although the direction of the relationship was surprising. Participants with lower working memory scores (low-spans) tended to attach relative clauses high, to NP1, and participants with higher working memory scores (high-spans) tended to attach low, to NP2. This trend held for both English and Dutch ← 91 | 92 → speakers, even though the overall attachment preferences differed between languages. In other words, consistent with previous research, English speakers still preferred NP2 attachment overall, and Dutch speakers still preferred NP1 attachment overall, but within each language, there were rather large systematic individual differences that exceeded those cross-linguistic differences. When we computed the effect size of language spoken on attachment preferences, we found that Cohen’s d = .29, which is regarded as a “small” effect. The effect size of the individual differences in attachment preference, on the other hand, was “large”, Cohen’s d = .72 in the English sample and .90 in the Dutch sample. To interpret these statistics a bit more, this means being a speaker of Dutch versus English accounts for about 30% of a standard deviation of the measure. But individual-specific verbal working memory score accounts for between 70–90% of a standard deviation. In short, individual differences in attachment preferences account for 3 times as much variance as cross-linguistic differences. One overall implication of this finding is that psycholinguistic processing principles once thought to be inflexible and automatic, such as Late Closure, can be shown to be highly flexible when examining individual differences. The second study in this line of research sought an explanation of the effects showing a NP2 preference for high-span participants. I will return to this study later to report those findings and their implications, but for now, I turn to another process once thought to be highly rigid and inflexible: sentence planning scope.
3. Variation in the scope of sentence planning
3.1. Inflexible units in sentence planning: A critical review
The most comprehensive examination of the functional language production system is Levelt’s model (Levelt, 1989; Bock and Levelt, 1994). It assumes a language production system with insular, sequential levels of processing: information at one level cannot be computed until receiving as input the output from the preceding level. First is the “message” level, the stage at which the basic semantic proposition the speaker intends to utter is composed. The next stage in processing is “grammatical encoding”, the accessing of non-phonological word information (meaning and syntactic category) plus the structuring of these so-called “lemmas” into their phrasal positions to produce surface structure. This surface structure representation is passed to ← 92 | 93 → “phonological encoding”, which retrieves word forms and creates a prosodic structure. Articulation is the final stage. A hallmark of this model is that each processing level works in parallel with the other levels in a pipe line mode. After the syntactic level outputs to the phonological level, syntax works on another piece of semantic input as phonology deals with the initial syntactic piece. Sentence production is termed incremental because of this parallelism and because as linguistic representations are shunted from one processing level to another, the range of operation is not over an entire sentence. Rather, speakers plan in increments, packaging small pieces of information together before that chunk is sent to the following level.
One aspect of this model that for a long time had been relatively uncontroversial is the claim that planning at many of these levels of representation is automatic: That is, such planning doesn’t require any processing resources such as working memory (Levelt, 1989; reviewed in Garrod and Pickering, 2007). Indeed, there would be obvious advantages found in a system that does not need cognitive resources to operate. However, in order to achieve that kind of freedom from resources, the planning system would have to sacrifice something: flexibility in the extent to which utterances can be planned in advance. Hence, such models assume that planning scope is stable, or inflexible, and architecturally minimal.
This assumption of automatic, inflexible units of planning has been accompanied by an empirical search for what those units might be. For example, Smith and Wheeldon (2001) conducted a set of experiments to test whether there is costly syntactic planning before speech onset. In their experiments, sentences like “The spoon and the car move up” were used to prime the production of syntactically related sentences (Experiments 1–5) in picture description tasks. When participants uttered sentences that were syntactically like the previous sentence, a reliable 50 ms advantage to begin speaking was found. The measure indicated how much time had been saved in the planning of the syntactic frame of the sentence. Smith and Wheeldon also tested the scope of this effect and found that it only held for the first phrase of an utterance, leading them to conclude that phrases are the automatically planned units of grammatical encoding. Another study (Griffin, 2001) showed that when speakers described scenes with 3 objects (at positions A, B and C) using sentences like “The A and the B are above the C”, only the word frequency of the object at position A influenced ← 93 | 94 → speech latency. Griffin concluded that speech planning is automatic, with a minimal, phrasal planning scope.
The notion that it will be possible to find fixed, automatic planning units is likely flawed for several reasons. As Levelt (1989) points out, in the time between 1967 and 1989, at least 18 different speech “planning units” (also sometimes referred to as “lookahead” or “scope”) of varying sizes had been proposed in the literature, leading Levelt to remark “…there is no single unit of talk” (Levelt, 1989, p. 23). Although this quote applies broadly across the different levels of representation listed above, there has since also been disagreement regarding what length those units might be, even within the same level of representation. For example, within the domain of grammatical encoding, although there are several studies that support the architecturally sub-clausal or phrasal view of incremental planning (Schriefers et al., 1998; Smith and Wheeldon, 1999, 2001), other studies show that the scope of planning extends to as much as a whole clause (Christianson and Ferreira, 2005; Ford and Holmes, 1978). Second, the studies that reveal minimal planning scopes neglect to apply pressures on planning processes to see whether the scope of planning can be flexibly pushed around in different situations. Lastly, there has been very little research into individual differences in planning scope. These reservations have given rise to an alternative account to rigid incremental planning during speech production: the flexible incrementality view. Much recent research has demonstrated that although there are circumstances when speech planning proceeds very incrementally—that is, bit by bit—there are also circumstances that dictate more planning to be done in advance (Costa and Caramazza; 2002; Damian and Dumay, 2007; Ferreira and Swets, 2002, 2005; Fuchs et al., 2013; Konopka and Meyer, 2010; Korvorst et al. 2006; Schriefers and Teruel, 1999a; Wagner et al., 2010). I will presently review some of these circumstances, including manipulations of time pressure and cognitive load.
3.2. Variation across situations
One way my colleagues and I have demonstrated this flexibility in planning scope is by having speakers produce sentences that can have some aspect of complexity manipulated very late in the sentence (e.g., Ferreira and Swets, 2002, 2005). In one such study (Ferreira and Swets, 2002), participants ← 94 | 95 → produced sentences as they solved simple math problems of subtly differing complexity. In the so-called “easy” condition, the speaker would see “21 + 22” on the screen, and the target utterance was “The answer is 43.” In the “hard” condition, the speaker would see “25 + 23” on the screen, and the target utterance was “The answer is 48.” Note that it takes people reliably longer to calculate the hard problem than the easy problem—sums totalling between 6 and 9 (such as the 8 in 48) take reliably longer to compute than sums totaling 5 or less (such as the 3 in 43) (Ashcraft, 1992). By exploiting this tendency, we estimated how far speakers planned their sentences in advance by measuring when speech slowed down in the hard condition relative to the easy condition. In Experiment 1, although we asked speakers to give their answers as quickly and as accurately as possible, the speakers were free to begin speaking whenever they chose. To assess the location at which speakers slowed down to plan for the difficult problems relative to the easy problems, we measured initiation time to begin speaking as well as the durations of the subsequent sections of the utterances, including each word of “The answer is” and both the 10s and 1s place of the arithmetic answer. Under these conditions, speakers slowed in the hard condition relative to the easy condition only in their latency to begin speaking. Once articulation began, we found no effects of problem difficulty on speech duration, implying that speakers had planned the entire sentence, including the solution to the addition problem, prior to the onset of speech.
In Experiment 2, we introduced an explicit deadline to begin speaking. During and after the practice phase of this experiment, a timer began counting down as soon as the arithmetic problem appeared on the screen. If participants did not respond before the timer finished, they heard a “beep” sound that indicated the deadline to begin speaking had passed. Under this deadline, effects of problem difficulty were found at each hand-measured section of an utterance, including the initiation time, “The answer is”, and the answer itself. In other words, even though speakers were doing some long-distance planning before beginning to speak, they were leaving some of the planning of these sentences for later. We concluded that the situation of time pressure had some influence on the scope of planning.
Since this research, several other studies (Damian and Dumay, 2007; Ferreira and Swets, 2005; Fuchs et al., 2013; Konopka and Meyer, 2010; Korvorst et al. 2006; Wagner et al., 2010) have demonstrated how planning ← 95 | 96 → scope can vary across situations. One noteworthy example of this was a series of experiments reported by Wagner et al. (2010), who found that increased task load reduces the scope of grammatical encoding. In that research, participants were either in the situation of having a low or high task load concurrently with the production task. The situation of increased load reduced planning scope. More recently, Fuchs et al. (2013) reported that different measures of planning reveal different simultaneous planning scopes. Whereas initial f0 peak was only sensitive to local planning considerations, measures such as pause duration, inhalation duration, and inhalation depth seemed sensitive to longer-distance planning effects.
3.3. Variation among individuals
This evidence that the scope of planning is flexibly adaptive to situational manipulations may indicate a planning system that is also adaptive to individual differences. Both the Wagner et al. (2010) and Fuchs et al. (2013) studies allude to the possibility that a large amount of variance in planning scope occurs among individuals. Fuchs et al. (2013) for example found large speaker-specific variation in planning as measured by breath inhalation depth prior to articulation. Whereas some of the participants would inhale very deeply prior to a long sentence compared to a short sentence (indicating a long scope of planning), some participants inhaled nearly equally in long and short sentences. Figure 1 illustrates this phenomenon, showing how the difference in inhalation between long and short sentences varies among the individual participants. The resulting figure illustrates a phenomenon that I mentioned at the outset, which is that most experimental studies are unable to account for variance among individuals. The data points in Figure 1 beg to be systematized and ordered on some dimension. As presented, these data points represent error variance: variance among participants that cannot be accounted for. But the goal of psychology, as mentioned earlier, is to find systematic variance—to find some way to straighten a seemingly random array of dots. Here is where correlational, individual differences techniques find their utility. ← 96 | 97 →
Figure 1. Results presented in Fuchs et al. (2013) illustrating the difference in inhalation depth between long and short sentences. Each point represents an individual participant. Reprinted from Journal of Phonetics, 41, Fuchs, S., Petrone, C., Krivokapić, J., and Hoole, P., Acoustic and respiratory evidence for utterance planning in German, 29–47, Copyright 2013, with permission from Elsevier.
One such correlational variable that can help systematically account for some of this variation among individuals, and “straighten out” figures like the one above, is planning time. Wagner et al. (2010), in an analysis done to help rule out alternative explanations of their results, found that the amount of time taken to begin articulating a given sentence reliably predicted the distance of the interference effects that indicated the scope of syntactic planning: The more quickly speakers initiated articulation, the less likely they were to show long-distance interference effects in their planning. More ← 97 | 98 → research has recently emerged showing similar patterns: If an individual speaker chooses a strategy of speaking sooner rather than later, there are associated reductions in planning scope (Gillespie and Pearlmutter, 2011; Lange and Laganaro, 2014). In one such experiment, Gillespie and Pearl-mutter (2011) elicited sentences that could potentially elicit subject-verb agreement errors such as The apple near the pies was/*were, and argued that increased numbers of errors suggests a longer scope of advance planning. Results showed that speakers with higher average speech onset time produced more such errors, indicating a longer scope of planning. Likewise, in an experiment that measured the scope of phonological planning by priming the first and second elements (noun-adjective or adjective-noun) in a sentence-initial noun phrase, Lange and Laganaro (2014) showed that only the participants who delayed the beginnings of their utterances showed priming beyond the first element. Such research suggests that those who take additional time tend to plan more material in advance of speech.
The other primary approach to examining planning scope variation among individuals has been to examine working memory capacity (Petrone et al., 2011; Swets et al., 2014). Although several previous studies employing a variety of approaches have demonstrated that higher-level language production, including grammatical encoding, is supported by working memory resources (Hartsuiker and Barkhuysen, 2006; Horton and Spieler, 2007; Kellogg et al., 2007; Kemper et al., 2003; Kemper and Sumner, 2001; Slevc, 2011), none of these studies had examined individual differences in planning scope. Petrone and colleagues (2011) found that working memory predicted the pitch of speakers’ voices to begin articulating phrases of different complexity, suggesting that speakers with more working memory may have a greater planning scope. The rationale of using initial utterance pitch as an indicator of planning scope is that longer phrases are associated with a greater pitch declination from start to finish. Speakers who can plan more in advance are those who are more likely to begin their sentences at a higher pitch to anticipate the upcoming declination. The results showed that high span speakers began articulation of complex subject phrases at a higher pitch than low span speakers. One interesting note about this finding to which I will return later is that despite this observed difference in apparent planning scope based on working memory, preparation time to begin articulation was equivalent among the groups (Petrone et al., 2011). ← 98 | 99 →
4. The role of working memory in planning
Speakers often utter sentences in circumstances of referential ambiguity, and in such circumstances, careful sentence planning can spell the difference between effective and ineffective communication. Suppose there is a carpenter’s assistant holding two hammers, and a harried carpenter who asks that assistant to “Hand me the hammer.” Had the carpenter planned more carefully, a more optimal sentence might be “Hand me the smaller hammer.” The hypothesis of a recent study I conducted with colleagues (Swets et al., 2014) was to investigate whether individual differences in working memory predict variation in the scope of advance sentence planning in such circumstances of referential ambiguity. We reasoned that someone with high working memory capacity might be capable of both gathering important information about ambiguities and integrating such information into their speech plans.
A secondary aim of the study was to identify what kind of role working memory plays in the planning process. One view of the role working memory might play in planning scope is that it affords a storage space for the messages one generates while planning. On this view, with limited working memory, only small increments can be planned at one time because a lower capacity prevents the storage of larger plans. According to this hypothesis, speakers with more working memory will plan more content in advance, but like the speakers in Gillespie and Pearlmutter (2011), Lange and Laganaro (2014), and Wagner et al. (2010), they should also have to spend more time creating those larger plans. We also tested an alternative view of the role of working memory that gives it not just storage functions, but also efficiency functions. According to this hypothesis, working memory performs the job not only of simple storage of generated message plans, but also of integrating and packaging linguistic information in a temporally efficient manner. As such, speakers with more working memory should be able to plan more of a sentence in advance, but do so without taking up additional time. Alternatively, they might plan the same amount as low-span speakers, but in less time (see Heitz and Engle, 2007 for a similar effect in the working memory literature). A prior result obtained in our lab had suggested this possibility of increased efficiency (Swets et al., 2013). In that study, we observed that speakers with co-present conversational partners provided more detailed ← 99 | 100 → descriptions of ambiguous objects, reflecting more careful planning, than speakers without a co-present addressee. However, the groups did not differ on initiation times. Perhaps participants with additional WM capacity are likewise capable of doing more planning on an equivalent time scale.
In order to test these hypotheses, we used eye tracking to measure the extent to which speakers inspected more advance regions of a visual display before beginning to describe it. The description paradigm we used was inspired by Brown-Schmidt and Tanenhaus (2006), who had previously used images containing two referents so similar to each other that to describe the picture to an interlocutor, one must distinguish between the two referents by using a modifying expression. We also measured individual differences in working memory via a reading span task (using the same materials as Swets et al., 2007). We hypothesized that working memory is used to prepare and store larger utterance plans, suggesting that people with high working memory capacity should literally look further ahead into the picture when planning high-level sentence information than people with low working memory capacity.
Figure 2 presents examples of the two types of displays we showed to participants in this study. In experimental conditions, we showed the cat with four legs in the first position, and the cat with three legs in the third position. Control conditions featured a different object in the third position, such as a wheel.
Figure 2. Examples of contrast and control displays in Swets et al. (2014)
The study consisted of two phases. We assessed working memory via reading span in phase I. In phase II, a sample of participants from phase I who demonstrated a wide range of reading span scores returned to act as Directors in a matching game. During the game, Directors produced utterances that mentioned the three objects and the directions of their movement (as ← 100 | 101 → indicated by the arrows). For example, in response to the contrast display, a participant might say, “The four-legged cat moves below the train and the three-legged cat moves above the train,” though certainly other descriptions that fit the target frame were possible.
Directors understood that the purpose of their utterances was to allow Matchers to manipulate items on a grid displayed on the Matchers’ own computer (see Figure 3). Because Matchers had the same cats in their displays as the Director, it was important for the Director to modify both the first noun, that corresponded to the object in Region 1, and the third noun, that corresponded to the object in Region 3 (see Figure 4). For control conditions, there was no need to modify either the first noun phrase (N1), cat, or the third noun phrase (N3), wheel, as the Matcher also saw only one cat in that condition.
Figure 4. Depictions of visual Regions 1 and 3, and noun phrases 1 and 3 (N1 and N3) for the contrast and control conditions of Swets et al. (2014).
Our dependent measures included initiation time, or the time taken to begin speaking, and fixation patterns, or where people were looking during particular windows of time. We also examined the content of N1 and N3 descriptions by coding whether participants modified N1 and N3. We ← 102 | 103 → treated working memory (WM) as a continuous measure to avoid artificial dichotomization. Results were analyzed using linear mixed effects models in R, with WM and display type entered as interactive fixed effects, and participants and items entered as random effects.
According to the general hypothesis that working memory supports advance planning processes, we predicted that we would observe correlations between working memory and our measures of advance planning. Specifically, working memory should correlate with the tendency to look at the contrast object in Region 3 (e.g., the three-legged cat) before speaking and with the tendency to modify N1 (e.g., The four-legged cat rather than The cat) early on in the sentence. Analysis of the time course in which these additional looks and modifications took place was intended to help distinguish between WM as a simple capacity limitation or a more active, efficient integration process. If high span participants require more time to plan additional content prior to speech, then WM could be viewed simply as storage for larger plans. If high span participants take the same (or less) time to plan more content than low spans, then the role of WM is more complex in that it also invokes temporally efficient packaging.
In an example of a typical observed trial, a Director with high reading span describing the cat display shown in Figure 4 might fixate both cat one and cat two before articulating a description that modifies both the first noun, N1, and the third noun, N3. A low-span individual describing the same display might fail to fixate the contrast cat in Region 3 before beginning to speak, and then fail to subsequently modify N1. You will see from these data that for contrast displays, high spans were more apt than low spans to not only fixate Region 3 before speaking, but also include a modification of N1 that helped listeners immediately distinguish between two possible referents.
The first analysis presented in Figure 5 is the amount of time Directors took to begin speaking. The figure shows that during contrast trials, working memory did not predict the amount of time to begin speaking: Everyone took about 2.5 seconds on average. On the other hand, during control trials, when there was less planning work to do, high spans began speaking more quickly than low spans. Consistent with the efficient capacity account, working memory allowed speakers in these control trials to plan equivalent content in less time. ← 103 | 104 →
Figure 5. Initiation time results from Swets et al. (2014). Reprinted from Language and Cognition, 6, Issue 01, 2014, 12–44, Swets, B., Jacovina, M.E., and Gerrig, R.J. Individual differences in the scope of speech planning: Evidence from eye movements. Copyright © 2014 UK Cognitive Linguistics Association. Reprinted with the permission of Cambridge University Press.
The next analysis examines speakers’ eye movements. Figure 6 shows the percent of this initiation time window that speakers spent looking at the object in Region 1 (e.g., the 4-legged cat). The results show that high spans spent less of their available time fixating Region 1 than low spans did if there was a contrast object in Region 3. WM did not correlate with this measure during control trials. This implies that high spans were less likely to be looking at Region 1 while preparing their utterances in the presence of a contrast. ← 104 | 105 →
Figure 6. Gaze duration results from Swets et al. (2014). The period of interest in the displayed results is the time between the appearance of the stimulus, and the onset of speech. The visual region of interest is Region 1, the object occupying the first (left-most) position in the display. Reprinted from Language and Cognition, 6, Issue 01, 2014, 12–44, Swets, B., Jacovina, M.E., and Gerrig, R.J. Individual differences in the scope of speech planning: Evidence from eye movements. Copyright © 2014 UK Cognitive Linguistics Association. Reprinted with the permission of Cambridge University Press.
The next analysis reveals that high span speakers spent this extra time that they otherwise would use to look at Region 1 by looking at the object in Region 3. Figure 7 shows the percent of initiation time that speakers spent looking at the object in Region 3 (e.g., the 3-legged cat in the contrast condition vs. the wheel in the control condition). In fact, during contrast trials, high span individuals were more likely to fixate the contrast objects in Region 3 prior to articulation than low spans. Neither high spans nor low spans tended to look at this region if there was no contrast to encode for description. ← 105 | 106 →
Figure 7. Gaze duration results from Swets et al. (2014). The period of interest in the displayed results is the time between the appearance of the stimulus, and the onset of speech. The visual region of interest is Region 3, the object occupying the third (right-most) position in the display. Reprinted from Language and Cognition, 6, Issue 01, 2014, 12–44, Swets, B., Jacovina, M.E., and Gerrig, R.J. Individual differences in the scope of speech planning: Evidence from eye movements. Copyright © 2014 UK Cognitive Linguistics Association. Reprinted with the permission of Cambridge University Press.
The next analysis will reveal whether the speakers with more working memory who were more likely to literally look ahead to Region 3 of the display also produced correspondingly detailed descriptions. Speakers in general created longer descriptions in the presence of a contrast, and those with more working memory were much more likely to do so by modifying the first noun. Figure 8 presents the likelihood that a speaker modified N1, either by calling the first cat a four-legged cat, cat with four legs, whole cat, or cat with two legs. The figure shows that working memory predicted this likelihood such that high spans were more likely to modify N1 than low-spans if there was a contrast. There was no such correlation during control trials. This result suggests that high spans are not only more apt to gather information about the contrast prior to speech, but also to encode that contrast very early on into their utterance plans. ← 106 | 107 →
Figure 8. Utterance type results from Swets et al. (2014) showing the likelihood that participants modified the first noun phrase in a target utterance (N1). Reprinted from Language and Cognition, 6, Issue 01, 2014, 12–44, Swets, B., Jacovina, M.E., and Gerrig, R.J. Individual differences in the scope of speech planning: Evidence from eye movements. Copyright © 2014 UK Cognitive Linguistics Association. Reprinted with the permission of Cambridge University Press.
To summarize, we found that reading span predicted speakers’ scope of planning. One of the first signs of these individual differences appeared during the initiation time window. Although high spans took the same amount of time to begin speaking as low spans when a contrast was present, they used that time much differently than low spans. Specifically, high spans spent more time gathering information about the similar objects to be distinguished for the matcher. This additional inspection of the more advance regions in the display allowed them to integrate information about the contrast earlier than low spans: They not only gave longer descriptions of N1, but also showed a greater likelihood of modifying N1 to verbalize the contrast with N3.
In support of the general hypothesis, we found that higher working memory capacity is associated with a larger scope of speech planning. It appears that speakers with high verbal working memory capacity are able to not only gather more information about a message before speaking, but ← 107 | 108 → also integrate that message early on in utterance plans. On the other hand, low spans are not as productive in using the time available to gather advance planning information. These results thus favor the efficient capacity view of working memory’s role in planning over the simple capacity view. Working memory capacity seems to allow speakers to be more efficient, or productive, in the extent to which they plan utterances in advance. High spans can not only plan more utterance information than low spans at a given time, but as Figure 5 shows, they created these larger plans without taking any additional time to do so. An account holding that WM simply stores larger utterance plans cannot account for such an effect. These effects are consistent with those observed by Petrone et al. (2011), who had found similar effects of efficient advance planning among individuals with high WM. Together these results argue for the role of working memory in sentence planning as a provider of efficient information integration in addition to a storage space.
Now that I have illustrated how the individual differences approach can help us to better understand the role of working memory in sentence production, I will now return to the domain of sentence comprehension. In doing so, I hope to illustrate that working memory’s role in both is quite similar.
5. The role of working memory in language processing
At the outset of the paper, I presented results from a study (Swets et al., 2007) showing that individual differences in working memory accounted for more systematic variance in relative clause attachment preferences than cross-linguistic differences, which suggests that working memory is involved in sentence comprehension in substantive ways. Then, I demonstrated that working memory plays a role in sentence planning, and illustrated that the nature of this role entails more than simple storage. The aim of this section is to show that individual differences approaches applied to multiple sentence processing domains can help explain the role that working memory might play more generally.
Recall from the earlier study (Swets et al., 2007) that individual differences in working memory capacity predicted ambiguous relative clause (RC) attachment such that high-spans attached low and low-spans attached ← 108 | 109 → high. We had not predicted this result. Our predictions had been based on assumptions regarding the mechanistic role that WM must play in parsing. Specifically, we had supposed that WM plays the role of simple storage: The more WM one had, the more likely they were to keep NP1 available in storage long enough to be associated with the RC. Given that we found the opposite result, we were forced to examine an alternative view of the role that working memory plays in this process.
To explain this finding, we proposed that WM plays a role that exceeds that of simple storage. Rather, it is a processing resource that allows comprehenders to chunk a certain amount of information together while reading. If high-span readers can “chunk” more information together while reading, they can regard the entire subject of the sentence as one “processing unit”. On the other hand, low-spans may have to break up the subject because of its length. A likely boundary for such a break point is just before the relative clause, which would separate the complex noun phrase (“The sister of the actress”) from the relative clause (“who shot herself on the balcony”). By placing such a mental boundary at that point, the NP1 could become a more appealing attachment site than NP2. Hence, our hypothesis was that chunking strategies underlie the individual differences observed in Study 1: Perhaps the reason low-span readers attach to NP1 is that they create smaller “processing chunks” as they read silently, leading to NP1 being the more viable attachment site. If this is true, then forcing all readers (including high spans) to use the same chunking strategies during reading should reduce the attachment preference differences between high- and low-spans.
Study 2 tested this hypothesis by forcing participants to parcel the complex NP and the RC into two pieces with an intervening break. Specifically, we presented each of the Study 1 sentences in 2-second chunks: first, the complex NP (“The maid of the princess”), followed by the relative clause with a modifying prepositional phrase PP (“who shot herself on the balcony”), then the matrix verb phrase VP (“was under investigation”). According to this hypothesis, if WM underlies the size of the processing chunks people use to parse syntax, then forcing a break between N2 and the RC should reduce or eliminate the relationship between WM and attachment preference by making everyone behave like low spans, whereby they attach high, to NP1. ← 109 | 110 →
Figure 9 reveals that chunking the text had precisely this effect on relative clause attachment preferences. The left side of the graph shows the Study 1 effects that were summarized earlier, and illustrates that participants with lower working memory were more likely to attach “high”, to NP1. The right side shows the results of Study 2, in which the text of the relative clause sentences was artificially chunked. The figure reveals that the relationship between Reading Span and RC attachment preferences was greatly reduced in English and apparently eliminated in Dutch. Also noteworthy is that percent NP1 attachments increased for all groups in both languages, and that English attachment and Dutch attachment both revealed an overall NP1 preference.
Figure 9. Overall results from Swets et al. (2007): Attachment preferences as a function of language (English vs. Dutch), presentation style (whole vs. chunked), and working memory category (low, mid and high span). ← 110 | 111 →
To sum up, in Study 1, the direction of the relationship between WM and attachment preference was the same in both English and Dutch: Individuals low in WM attached high, and individuals high in WM attached low. In Study 2, chunking the text reduced the relationship between WM and attachment preference significantly because it effectively forced everyone to adopt the chunking strategies used by low spans. From this set of results, we can draw two conclusions. First, the final products of parsing are bounded by the limits of working memory capacity. But more germane to the present argument, we can also conclude that the mechanistic role that working memory plays in this parsing process is not simply to store potential attachment sites. Rather, working memory is functioning at a high level during which initial packages of information are assembled together. The more working memory one has, the more likely one is to assemble large packages of information to be analyzed for later parsing decisions.
By examining this study in concert with the study on individual differences in planning scope, we can also draw a conclusion about the role of working memory in language processing more generally: Working memory predicts informational chunking not just in the manner in which we plan our sentences during language production (Swets et al., 2014), but also in aspects of language comprehension. Moreover, its role cannot be reduced to that of simple storage.
Working memory’s role was similar in both domains: In the sentence planning study (Swets et al., 2014), participants were able to create larger utterance plans with additional WM; and in the attachment preferences study, participants were able to package more linguistic material together for parsing/analysis with additional WM. Taken together, this suggests that in perhaps all domains of sentence processing, working memory plays the role of packaging information together for both purposes of storage and active, efficient integration with other information. Furthermore, any differences that can occur in working memory capacity be they individual differences or experimentally manipulated reductions in capacity, by influencing the size of these packages and the efficiency with which they are assembled, can influence the basic mechanisms of sentence processing. I am currently collecting data from other domains to determine whether this principle of working memory as information packager applies even more generally. One such area of investigation is in the use of lexical and event information ← 111 | 112 → to predict upcoming elements in a sentence (Altmann and Kamide, 1999), which could show that the scope of prediction, like the scope of planning or the size of informational packages assembled during parsing, varies along with working memory.
6. Future directions
So far, I have summarized some previous research demonstrating that individual differences research can help explain variance in the scope of planning in language production. Although this early research is promising, there is still more work that can be done in this research vein to help understand the cognitive architecture associated with planning in language production. Such future directions of this research ought to include additional individual differences measures, examine more levels of planning, and compare individual differences to cross-linguistic differences in planning scope.
More individual differences measures. To this point, the only individual differences measures that have been shown to correlate with the scope of advance speech planning are working memory (Petrone et al., 2011; Swets et al., 2014) and preparation time (Gillespie and Pearlmutter, 2011; Lange and Laganaro, 2014; Wagner et al., 2010). But WM is known to correlate with other aspects of cognition such as attentional control (Kane et al., 2004) and processing speed (Salthouse, 1994). To complicate things even further, there are multiple aspects of working memory, including possible systems that serve strictly verbal WM, strictly spatial WM, and a more general WM that underlies all cognitive processing (Swets et al., 2007). The effects of WM on planning processes may represent modest advances, but they also do not account for enough variance to consider the case closed. In other words, there are more sources of individual differences to pursue, including for example age, attentional control, and personality factors.
Two sources of potential systematic variance that seem especially ripe for further investigation are processing speed and speech rate. It is possible that individual differences in processing speed might account for the observed link between working memory and advance sentence planning. Prior research (summarized in Salthouse, 1994) has documented that age-related declines in working memory can be largely attributed to declines in ← 112 | 113 → processing speed, which points toward a possible processing speed explanation of our WM findings. Furthermore, it is also possible that processing speed, the rate at which speakers articulate linguistic material, and planning might be interrelated in ways that to this point have not been explored in the literature. Hence, it will be important in future research to include measures of processing speed separable from working memory capacity to sort out how these various facets of cognitive performance help facilitate the speed, fluency, and scope of planning in language production.
Of course, the search for other individual differences factors that might be associated with the planning scope of language production need not be limited to just the above factors. For example, in other domains of language production such as syntactic priming, measures of individual differences including Big Five personality factors such as extraversion and conscientiousness (Gill et al., 2004), age (Kidd, 2012), and perspective-taking and autism (Horton, 2014) have been found to correlate with the extent to which a speaker will re-use a syntactic structure just uttered by an interlocutor. Perhaps such factors could explain meaningful variance in planning scope as well. For example, Horton (2014) had found that individuals higher in perspective-taking ability were more likely to align with their partner on the type of syntactic structure they produced in a picture description task. In other words, individuals who had been measured to be highly sensitive to their partner’s perspectives produced language that also seemed sensitive to the perspectives of their partners. Given that an essential element of proper advance planning in the picture description task described in Swets et al. (2014) is understanding the communicative needs of the addressee, it is possible that individuals higher in perspective-taking would also be more likely to plan more of their sentence in advance of articulation.
One other individual differences measure that deserves some mention in this context is something known as the BLIRT measure (Brief Loquaciousness and Interpersonal Responsiveness Test, Swann and Rentfrow, 2001). “Blirtatiousness”, defined by Swann and Rentfrow (2001) as “how quickly, frequently, and effusively people respond to their partners” could be used as a measure to variance related to personality, perspective-taking, and temporal factors related to the planning of language. Those high in “blirtatiousness” talk quickly and often, and those low in the factor are more measured in their speech output. Use of this scale in future research ← 113 | 114 → might capture a great deal of variance in planning scope among speakers, but also risks being too general a measure to explain specific mechanisms.
Levels of planning. An additional limitation of the research so far is that measures of WM have only been shown to correlate with two aspects of speech planning: prosodic plans (Petrone et al., 2011) and the interface of message level planning and utterance planning (Swets et al., 2014). But limitations of this previous research invite more work to be done. For example, one limitation of the research on individual differences in prosodic planning (Fuchs et al., 2013; Petrone et al., 2011) is that it is based on read speech. However, in line with Swets et al. (2014), it would be preferable to examine speech planning in more interactive situations, when the linguistic content is generated primarily by the speaker.
But the primary limitation of this previous research is that it has considered only a small sample of the range of representations that are planned during language production. As mentioned above, speech must be planned at several levels of representation (Levelt, 1989). With only message level and prosodic level representations considered so far in the study of individual differences in planning scope (Petrone et al., 2011, Swets et al., 2014), other levels to be considered include phonological encoding, grammatical encoding, and lemma selection. Hence, future projects ought to examine the planning scope of language production at multiple levels of representation as a function of multiple measures of individual differences.
A cross-linguistic approach. Similar to the research line taken in Swets et al. (2007), one intriguing direction of individual differences research in the domain of language production might be to compare effects of individual differences on speech planning across multiple languages. Although some previous research (Janssen, et al., 2008; Brown-Schmidt and Konopka, 2008, Christianson and Ferreira, 2005; Schriefers and Teruel, 1999b; see Jaeger and Norcliffe, 2009, for a review) has investigated differences in incremental planning between languages, and some more recent research has focused on individual differences, no previous research on the flexibility of planning scope has ever simultaneously compared planning scope differences among various languages to planning scope differences among individuals of the same language community. It would be interesting to examine how the cross-linguistic differences in grammatical encoding scope found by Schriefers and Teruel (1999b), for example, would compare in effect size ← 114 | 115 → to individual differences in the scope that is predicted by WM. Cataloging the factors that influence planning scope within and across languages can help researchers map the contours of the language production system that all languages share.
Notes of caution. I hope that this chapter conveys the enthusiasm I have for individual differences approaches in studies of speech production and perception. But before I conclude, it seems appropriate to give some words of caution regarding their utility. First, let me repeat the most important warning: Because individual differences research is inherently correlational, one cannot claim a causal link between an individual differences measure and performance on some linguistic task. Fortunately, there are ways to address this limitation of the approach. First, although one cannot assume a causal link when finding a significant correlational relationship, correlation also does not rule out the existence of such a link. So perhaps it is sometimes better to regard a significant correlation between, say, working memory and planning scope, not as evidence that increased capacity causes increased planning scope, but as initial evidence for such a link which must be confirmed by subsequent research. To find support for such a causal link, one could conduct experiments by manipulating a task circumstance that is thought to use resources that are associated with the individual differences measure that has already been shown to correlate with a linguistic measure of interest. For example, if we are considering working memory, manipulation of task load as in Wagner et al. (2010) can simulate the high- or low-span functionality that is naturally expressed in individual differences. A more exploratory, but intriguing technique to help confirm causal explanations of individual differences findings is tDCS (trans-cranial direct current stimulation). With tDCS, a researcher places anode and cathode electrodes at different sites on the scalp and delivers a low-voltage current through the head. By placing the anode electrode over a region of interest, one can “stimulate” that region. One way this technique has been used in language studies has been to show that placing an anodal electrode over the prefrontal cortex (PFC), associated with executive function, participants planned sentences that were less error-prone than participants who received a control (sham) stimulation condition (Nozari, et al., 2014). Grammatical fluency in conversational speech has also improved by placing the anode over Broca’s area (Marangolo et al., 2013). One appealing aspect of tDCS ← 115 | 116 → over task manipulation in general is that one can attempt to isolate and improve the cognitive process (WM, executive function, grammar) that is thought to explain the phenomenon under investigation rather than develop some task that merely disrupts task performance.
Beyond the interpretational limitations of correlational research, studies of individual differences tend to require larger numbers of participants than traditional experimental designs. The number required follows from the kinds of questions one wants to address. For example, in our study on individual differences in relative clause attachment (Swets et al., 2007), we were testing hypotheses for which we needed to tease apart different sources of working memory variance to see how they each predicted relative clause attachment preferences. To do this, we required two types of statistical analyses that are quite greedy regarding numbers of participants: factor analysis and structural equation modeling. Our large samples ranging from 96 to 150 participants per study proved to be manageable in a study of sentence comprehension with easily codable binary responses, but for studies of language production with their large demands on time for transcription and coding, such numbers of participants can prove to be intractable. Our dual-solution to this problem in the working memory/planning scope research (Swets et al., 2014) was to perform simpler statistical analyses using mixed models in R, and to “rig” the working memory sample to ensure greater variability. Although we initially collected working memory data from nearly 100 participants, we invited only 26 participants back to participate as speakers in the picture description task. The 26 who returned showed a wide range of working memory scores because we invited back everyone who scored in the more extreme ends of the distribution, and sent fewer invitations to participants in the center. By maximizing the variance in working memory, we were able to find a significant relationship between planning scope and working memory among this small number of participants.
This final warning is one that would seem obvious, but still warrants a mention. Of course every researcher understands that it is inadvisable to conduct unmotivated studies. However, this advice is far easier to heed when the studies one conducts involves experimentally manipulated variables that one must counterbalance and fuss over before any data are collected. The temptation when dealing with individual differences measures ← 116 | 117 → is to include them in any study one performs, without necessarily justifying their inclusion. Although individual differences measures have great potential to bring language production researchers an increased understanding of variables related to speech planning, we must be judicious in using them, and provide sound theoretical motivation for every measure we take.
Altmann, G. T. M. and Kamide, Y. (1999). Incremental interpretation at verbs: Restricting the domain of subsequent reference. Cognition, 73, 247–264.
Ashcraft, M. J. (1992). Cognitive arithmetic: A review of data and theory. Cognition, 44, 75–106.
Bock, J. K., and Levelt, W. J. M. (1994). Language production: Grammatical encoding. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics (pp. 945–984). San Diego: Academic Press.
Brown-Schmidt, S. and Konopka, A. E. (2008). Little houses and casas pequeñas: message formulation and syntactic form in unscripted speech with speakers of English and Spanish. Cognition, 109, 274–280.
Brown-Schmidt, S., and Tanenhaus, M. (2006). Watching the eyes when talking about size: An investigation of message formulation and utterance planning. Journal of Memory and Language, 54, 592–609.
Brysbaert, M., and Mitchell, D. C. (1996). Modifier attachment in sentences parsing: Evidence from Dutch. Quarterly Journal of Experimental Psychology, 49, 664–695.
Christianson, K. and Ferreira, F. (2005). Planning in sentence production: Evidence from a free word-order language (Odawa). Cognition, 98, 105–135.
Costa, A., and Caramazza, A. (2002). The production of noun phrases in English and Spanish: Implications for the scope of phonological encoding in speech production. Journal of Memory and Language, 46, 178–198.
Daneman, M., and Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466.
Damian, M. F., and Dumay, N. (2007). Time pressure and phonological advance planning in spoken production. Journal of Memory and Language, 57, 195–209.
Ferreira, F., and Swets, B. (2002). How incremental is language production? Evidence from the production of utterances requiring the computation of arithmetic sums. Journal of Memory & Language, 46, 57–84.
Ferreira, F., and Swets, B. (2005). The production and comprehension of resumptive pronouns in relative clause ‘island’ contexts. In: Cutler, A. (Ed.), Twenty-first Century Psycholinguistics: Four Cornerstones (pp. 263–278). New Jersey: Lawrence Erlbaum Associates Publishers.
Ford, M., and Holmes, V. M. (1978). Planning units and syntax in sentence production. Cognition, 6, 35–53.
Frazier, L. (1979). On Comprehending Sentences: Syntactic Parsing Strategies. Doctoral dissertation, University of Connecticut.
Frazier, L. (1987). Theories of sentence processing. In L. J. Garfield (Ed.), Modularity in knowledge representation and natural-language understanding. Cambridge, MA: MIT Press.
Fuchs, S., Petrone, C., Krivokapic, J., and Hoole, P. (2013). Acoustic and respiratory evidence for utterance planning in German. Journal of Phonetics, 41, 29–47.
Garrod, S., and Pickering, M. J. (2007). Automaticity of language production in monologue and dialogue. In A. S. Meyer, L. R. Wheeldon, A. Krott, (Eds.), Automaticity and control in language processing (pp. 1–20). New York, NY US: Psychology Press.
Gill, A. J., Harrison, A. J., and Oberlander, J. (2004). Interpersonality: Individual differences and interpersonal priming. Proceedings of the 26th Annual Meeting of the Cognitive Science Society (pp. 464–469). Mahwah, NJ: Erlbaum.
Gillespie, M. and Pearlmutter, N.J. (2011). Effects of semantic integration and advance planning on grammatical encoding in sentence production. In L. Carlson, C. Holscher, T. Shipley (Eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society (pp. 1625–1630). Austin, TX: Cognitive Science Society. ← 118 | 119 →
Griffin, Z. M. (2001). Gaze durations during speech reflect word selection and phonological encoding. Cognition, 82, B1–B14.
Hartsuiker, R. J., and Barkhuysen, P. N. (2006). Language production and working memory: The case of subject-verb agreement. Language and Cognitive Processes, 21, 181–204.
Horton, W. S. (2014). Individual differences in perspective taking and field-independence mediate structural persistence in dialog. Acta Psychologica, 150, 41–48.
Horton, W. S., and Spieler, D. H. (2007). Age-related differences in communication and audience design. Psychology and Aging, 22, 281–290.
Jaeger, T. F. and Norcliffe, E. (2009). The cross-linguistic study of sentence production. Language and Linguistics Compass, 3, 866–887.
Janssen, N., Alario, F.-X, and Caramazza, A. (2008). A word-order constraint on phonological activation. Psychological Science, 19, 216–220.
Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., and Engle, R. W. (2004). The generality of working memory capacity: A latent-variable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 133, 189–217.
Kellogg, R. T., Oliver, T., and Piolat, A. (2007). Verbal, visual, and spatial working memory in written language production. Acta Psychologica, 124, 382–397.
Kemper, S., Herman, R. E., and Lian, C. H. T. (2003). The costs of doing two things at once for young and older adults: Talking while walking, finger tapping, and ignoring speech or noise. Psychology and Aging, 18, 181–192.
Kemper, S., and Sumner, A. (2001). The structure of verbal abilities in young and older adults. Psychology and Aging, 16, 312–322.
Kidd, E. (2012). Individual differences in syntactic priming in language acquisition. Applied Psycholinguistics, 33, 393–418.
Konopka, A. E., and Meyer, A. S. (2010). Looking ahead: Variability in planning scope for complex noun phrases – evidence from eye-tracking [Abstract]. In Proceedings of the 16th Annual Conference on Architectures and Mechanisms for Language Processing [AMLaP 2010] (pp. 177). York: University of York. ← 119 | 120 →
Korvorst, M., Roelofs, A., and Levelt, W. J. M. (2006). Incrementality in naming and reading complex numerals: Evidence from eyetracking. The Quarterly Journal of Experimental Psychology, 59, 296–311.
Lange, M.V., and Laganaro, M. (2014). Inter-subject variability modulates phonological advance planning in the production of adjective-noun phrases. Frontiers in Psychology, 5, 43.
Levelt, W. J. M. (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press.
Marangolo, P., Fiori, V., Calpagnano, M. A., Campana, S., Razzano, C., Caltagirone, C., and Marini, A. (2013). tDCS Over the left inferior frontal cortex improves speech production in aphasia. Frontiers in Human Neuroscience, 7, 539.
Nozari, N., Arnold, J. E., and Thompson-Schill., S. L. (2014). The effects of anodal stimulation of the left prefrontal cortex on sentence production. Brain Stimulation, 7(6), 784–792.
Petrone, C., Fuchs, S., and Krivokapić, J. (2011). Consequences of working memory differences and phrasal length on pause duration and fundamental frequency. Paper presented at the Proceedings of the 9th International Seminar on Speech Production (ISSP), 393–400. Montréal, Canada.
Salthouse, T. A. (1994). The aging of working memory. Neuropsychology, 8, 535–543.
Schriefers, H., and Teruel, E. (1999a). Phonological facilitation in the production of two-word utterances. European Journal of Cognitive Psychology, 11, 17–50.
Schriefers, H., and Teruel, E. (1999b). The production of noun phrases: a cross-linguistic comparison of French and German. In Proceedings of the Twenty-First Annual Conference of the Cognitive Science Society, (Mahwah, NJ: Lawrence Erlbaum), 637–642.
Schriefers, H., Teruel, E. and Meinshausen, R. M. (1998). Producing simple sentences: results from picture-word interference experiments. Journal of Memory & Language, 39, 609–632.
Slevc, L. (2011). Saying what’s on your mind: Working memory effects on sentence production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 37, 1503–1514.
Smith, M., and Wheeldon, L. (2001). Syntactic priming in spoken sentence production—an online study. Cognition, 78, 123–164.
Swann, W. B. Jr. and Rentfrow, P. J. (2001). Blirtatiousness: Cognitive, behavioral, and physiological consequences of rapid responding. Journal of Personality and Social Psychology. 6, 1160–1175.
Swets, B., Desmet, T., Hambrick, D. Z., and Ferreira, F. (2007). The role of working memory in syntactic ambiguity resolution: A psychometric approach. Journal of Experimental Psychology: General, 136, 64–81.
Swets, B., Jacovina, M. E., and Gerrig, R. J. (2013). Effects of conversational pressures on speech planning. Discourse Processes, 50, 23–51.
Swets, B., Jacovina, M. E., and Gerrig, R. J. (2014). Individual differences in the scope of speech planning: Evidence from eye movements. Language and Cognition, 6, 12–44.
Wagner, V., Jescheniak, J. D., and Schriefers, H. (2010). On the flexibility of grammatical advance planning during sentence production: Effects of cognitive load on multiple lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36, 423–440. ← 121 | 122 → ← 122 | 123 →